=Paper=
{{Paper
|id=Vol-2370/paper-05
|storemode=property
|title=Cross-domain Research Data Management with Linked Data technologies
|pdfUrl=https://ceur-ws.org/Vol-2370/paper-05.pdf
|volume=Vol-2370
|authors=André Langer
|dblpUrl=https://dblp.org/rec/conf/caise/Langer19
}}
==Cross-domain Research Data Management with Linked Data technologies==
https://ceur-ws.org/Vol-2370/paper-05.pdf
PIROL: Cross-domain Research Data Publishing
with Linked Data technologies
André Langer[0000−0001−7073−5377]?
Chemnitz University of Technology, Germany
andre.langer@informatik.tu-chemnitz.de
Abstract. Effective research data management for traceability, preser-
vation and reuse is an important part of good scientific practice and is
already under discussion over a long period of time. However, the digital
transformation in science also led to new challenges for researchers on
how to describe, publish and share their research data. This includes the
interdisciplinary annotation and discovery of research data, data privacy
issues in exposure of data with trends to decentralized platforms as well as
sophisticated automatisms to ensure data quality and compliance aspects.
Only limited tool support exists for these processes so far.
The following research project will use Linked Data principles to improve
the current situation in this problem domain. It will first focus on com-
ponents and services, that assist researchers in the annotation process of
their research data. Next, it will investigate how this research data can
be stored and discovered in decentralized, multi-user scenarios to allow
data reuse under respect of data privacy concerns. In a third step, meta
data descriptions will be used to apply automated data conformance and
quality assessment operations on scientific data.
Keywords: Data Annotation, Data Publishing, Decentralization, Data
Quality, Linked Data, SoLiD, Open Research
1 Introduction
The term digital research data (sometimes also referred to as scientific
data or scholarly data) covers in principle any kind of digital
artifact that is associated with scientific research [9].
Research data is an essential artifact of scientific work: It leads to insights,
makes research reproducible and validates findings. Therefore, it is inseparable
connected to the results of research and also forms the base for future activities.
The digital transformation of science has risen new opportunities and issues for
all involved stakeholders on national and international level on how to annotate,
publish, archive, find and reuse digital research data. Researchers are confronted
with new challenges when aiming to publish and share or reuse research data digi-
tally. This applies especially for multifaceted research data where researchers from
many cross-domain knowledge domains work together in a collaborative fashion.
?
This research is developed under the supervision of Prof. Dr.-Ing. Martin Gaedke.
The process of research data management is commonly described in a data life-
cycle. Griffin et al. present a researcher-focused approach and describe five phases
for the complex process of research data management, which affect each other. Sev-
eral modifications and interpretations of this life-cycle can be found. Differences
primarily focus on the introduction of a separate planning phase for research data,
between the discovery of existing data and the generation of new research data as
depicted in fig. 1. Within this data life-cycle, the intended PIROL research project
will focus on the publication process (6) of research data. This will obviously also
be influenced by the storage of data and have an impact on the discovery.
Fig. 1. Research Data Management Life-cycle
Common online publication channels reach from direct data exchange, tradi-
tional self-hosted webserver downloads, up to larger centralized platform infras-
tructures and public/private cloud-based solutions with corresponding application
providers. Projects such as Google’s Dataset Search, OpenAIRE or Re3Data.org
show, that there are already several data publication platforms that provide
datasets with basic meta descriptions. Many of them also consider the usage of
semantic meta data descriptions, persistent identifiers and concept URIs. Never-
theless, ongoing activities in projects such as RADAR or the European Open
Science Cloud (EOSC) show, that no all-encompassing solution has been accepted
so far and that there is still a need for further research in this domain.
The PIROL research project will investigate the following research ques-
tion: How can cross-domain research data publishing processes be facilitated,
validated and extended for interdisciplinary reuse? Cross-domain means, that
it focuses on interdisciplinary research data management, where data shall be
made findable, accessible and reusable over knowledge domain specific bound-
aries on a meta level, so that researchers and institutions from similar and
other disciplines can benefit from it. But it also refers to the nature of domain-
specific data silos, and the desire to publish data in an institutional- and
platform-independent decentralized fashion.
The rest of the paper discusses the scientific approach to enhance digital
research data publishing in a hyper-connected society in the following way:
Section 2 describes existing problems and outlines related solutions and the
current status of these challenges. Section 3 defines in a concise way the main
objectives of the intended PIROL research project. Already existing results and
preliminary work from our early stage PhD experience is presented in section 4
before the overall conceptual approach and planned research methodology is
defined in section 5. The intended contribution is explicitly listed in section 6
and section 7 summarizes the project scope and limitations.
44
2 Problem Analysis
We focus on the following three late-breaking challenges in the
publishing process for digital research data:
Insufficient data annotation Researchers are not necessarily aware of how to
annotate their research data correctly for an interdisciplinary reuse.
Centralized data publishing Researchers commonly have to upload their
data on an institutional or even commercial centralized platform.
Weak data validation An assessment of published research data and meta
data is often done manually, crowd-based or even not at all. (Semi-)automated
reusable processes are still missing.
2.1 Insufficient data annotation
Annotating research artifacts is common and especially in the submission and
archiving context of scientific publications well-established. Complex classifica-
tion systems and taxonomies already exist as well as domain-specific controlled
vocabularies with different terms and conventions. The increasing role of knowl-
edge graph in the WWW also introduced alternative approaches to classify and
link data. The Semantic Web community proposed a Linked Data approach [1]
for that. Unique concept identifiers play an important role and are still under
scientific investigation in an interdisciplinary context123 .
However, to find datasets for a certain research problem is still difficult,
especially in an interdisciplinary context. Provided information primarily focuses
on discovery meta data, thus provenance information, and less on structured
data describing the content of the dataset and other aspects such as access
and legal information in a homogeneous fashion.
Furthermore, the annotation process itself seems to be tedious for the majority
of users, cognitively overloads them with the large amount of information to
provide, and results in sparely provided, narrow-minded meta information with
ambiguous content in the end. Ontologies such as schema.org/Dataset can act as
a blueprint, but it remains hard especially for inexperienced and non-technical
users to select relevant properties and manually annotate a dataset based on these
vocabularies; especially in the context of Linked Data. Appropriate components,
services and tools in the data management life-cycle could reduce this barrier for a
researcher, but it appears that no solution was widely accepted so far, or is bound
to a particular data management platform. This especially applies to frontend user
interfaces for Linked Data input. Research constantly concentrated on ontology
design, but then stopped with the manual creation of meta data information in
an RDF serialization format or through simple tabular input interfaces. Only
sparse research contributions can be found for frontend components for Linked
Data [3] or even a human-centered design approach.
1
https://www.project-freya.eu/en/about/mission
2
https://project-thor.eu/
3
https://2019.semantics.cc/robust-identifiers-leading-quality-data
45
2.2 Centralized data publishing
Platforms to publish research data encompass institutional-specific infrastructure
solutions, Open Data Platforms (CKAN, DuraSpace, Dataverse, EPrints, iRODS,
Invenio and others), advanced platforms (Dryad, Zenodo, EUDAT, Figshare)
as well as commercial providers such as ResearchGate and Mendeley or even
general collaboration and data sharing platforms such as Github, Google Drive,
Dropbox and further more [10]. But surveys such as the Wiley Open Science
Researcher Insights4 show, that certain researchers hesitate to use centralized
platforms and want to retain control on their research data, especially wrt. data
privacy concerns and existing legal agreements. This is reasonable especially in
the ongoing debate on data access, protection and data ownership. Decentralized
approaches could lower this barrier, increase transparency and eliminate data
silos and replicas. Solutions such as e.g., DatProject5 or EUDAT B2/EOSC6 )
already exist but it appears that no solution was widely accepted so far.
A Linked Data platform (LDP) based approach could even more intercon-
nect research data and its meta data from even heterogeneous platforms in
a unique fashion. SoLiD server implementations were already suggested and
successfully applied in different scenarios [8], but to the best of our knowledge
not tested for research data use cases expect with LDP platform implementa-
tions such as Fedora. Further open research challenges exist in how to specify
cross-platform access rules and how to run queries among multiple data stores
by considering privacy and latency aspects, e.g., by relying on Link Traver-
sal Based Query Execution (LTBQE) [11]
2.3 Weak data validation
High-quality input data is one of several basic requirements for successful business
operations because it directly affects consecutive process results in established
business value chains. This is especially true for the discovery of published
research data in a cross-domain context. However, assuring data quality for
published research data and meta data in data repositories is not trivial and
often involves human interaction, tedious reviews, or is not done at all.
When research data is published in data repositories together with correspond-
ing meta data information, automatic data conformance and data quality checks
can be run. In the past, research primarily focused on reasonable standard metrics
that could be used and measured as an indicator for data quality. A survey in
2014 identified 18 appropriate quality dimensions with 69 different metrics from
118 related articles [12] for assessing Linked Data quality. Depending on the type
of data source, they allow conformance measurements on data instance level,
ontological schema level as well as on service level. It is still a challenge to transfer
these data quality metrics to automated data quality assessment processes.
4
https://www.wiley.com/network/researchers/licensing-and-open-access/
open-science-trends-you-need-to-know-about
5
https://datproject.org/
6
https://www.eosc-hub.eu/
46
3 Objectives
To contribute to the systematic, efficient and sustainable cross-domain publication
process of research data, PIROL is going to accomplish the following objectives.
O.1 Simplify the annotation process for research data, especially for non-technical
users in an interdisciplinary context
O.2 Enable decentralized research data publishing activities on independent
platforms for research data
O.3 Support automatic assessment operations on provided research information
4 Preliminary Results
In order to achieve the defined objectives in the problem domain of research
data management, expertise in information management, semantic technolo-
gies and Web Engineering is needed. The first year of this early-stage PhD
project focused on the familiarization with the State of the Art in Linked Data
technologies. In the following, we list already achieved preliminary result that
will already contribute to the PIROL project. A national growth-core project
on Linked Enterprise Data Services7 in Germany funded by the BMBF sup-
ported this initial phase. The primary focus was set on data coherence and data
quality assessment operations for general-purpose Linked Data sets accessible
via existing web services or placed in a data lake.
A first result was a discussion of definitions on data quality and the deduction
of a formalized expression and recommendation on how to compute data quality
values for data sources based on multiple criteria on instance, schema and service
level (FAME.Q [4]). In a second step, a data quality assessment component
for Linked Data sources was implemented that is capable of measuring a basic
set of 55 recommended intrinsic, representational, contextual and accessibility
quality metrics based on open standards such as W3C’s RDF, SPARQL and
DQV (SemQuire [7].)A third preliminary result was the design of an ontology
that is capable of specifying requirements on data quality characteristics in a
uniform fashion to fill a gap between discovery and measurement activities. It
can be used for data quality assessment purposes to compare multiple eligible
data resources on particular metrics and attributes of current interest. This
formalization includes desired measurement result boundaries, a customizable
calculation as well as comprehensible quality rating scores (DaQAR [5])
Another research activity was already conducted in the context of building
knowledge graphs on interdisciplinary scientific publications. It became obvious,
that lots of contributions already exist in the ontological description and backend
implementation of Linked Data applications, but that user interfaces require a
user to deal with URIs directly. We therefore conducted a user study on how this
user interface experience can be improved especially for data input operations and
suggested a component-based approach on how to hide Linked Data identifiers
based on auto-suggestion features (URI-aware UI [6]).
7
http://leds-projekt.de
47
5 Approach
Project PIROL: Publishing Interdisciplinary Research Over Linked data will
extend our preliminary results and focuses on a customer journey in the narrowed
problem domain of research data management as depicted in fig. 2
Fig. 2. SAVE Research Data Publishing Journey
We introduce four stages for research data publishing and call them SAVE
steps that are aligned to the objectives from section 3. The project will not
discuss in detail the Store step with underlying devices and archiving solutions
for research data on a physical level. But we will consider persistent versioning
aspects and a Linking between derived or related datasets. Future work to
enhance these stages is briefly described in the following sections.
5.1 Simplify the annotation process for research data
Research in this section will contribute to the Annotation stage in fig. 2. In a first
step, we are going to compare concepts in existing vocabularies on an ontological
level with automatic clustering mechanisms. We have the hypothesis that many
researchers are not aware of what they should provide as meta information or are
not clear which vocabulary to use for annotation purposes. Ontologies such as
the schema.org Dataset vocabulary or DataCite are promising approaches, where
our approach can contribute recommendations for extension and modification.
In a second step, we investigate appropriate concepts for research data de-
scription. Altough Linked Data sources for general-purpose content already
exist (examples are DBpedia and Wikidata), this is not necessarily the case for
research-specific terms and concepts. Therefore, we will exemplarily introduce an
extensible registry for entity URIs to also describe concepts such as a particular
measurement method, evaluation method, metric or other characteristic. We
will check in a pre-analysis, if existing system platforms such as Wiki-based
Linked Data platforms can be used for that purpose.
In a third step, we will conduct research on alternative user input interfaces
that assist a user in the process of providing relevant meta data as conceptu-
ally shown in fig. 3. This can encompass intelligent data input components,
a preprocessing of the provided research data together with an AI-supported
classification, as well as context-aware faceted UI approaches.
48
5.2 Enable decentralized research data publishing activities
Research in this section will contribute to the Exposure stage in fig. 2. In a
preparation phase, we will first investigate the current status of existing and
established (open) data management platforms, to which extent they support
Linked Data references and data privacy aspects. We will then focus on a
decentralized LDP approach and apply recent SoLiD considerations on research
data management. Following the Linked Data paradigm, a user can place its
research data (together with a meta description) on any online platform, where
the content is basically accessible and referenceable with a URI.
Therefore, we will develop a concept on how to extend existing platforms
with a decentralized SoLiD LDP extension as well as how to run basic queries
in this environment. When searching research data with particular character-
istics, we are especially interested in how to formulate corresponding queries
and how to compare multiple query results. Additionally, approaches on how
to actively communicate changes in published research data to research data
catalogs in an automated fashion can be discussed.
5.3 Support automatic assessment operations
Research in this section will contribute to the Validation stage in fig. 2. We will put
special emphasis on the automated validation operations in the publishing step
for research data as we will have a corresponding profile with meta information
that can be used for assessment purposes. Post-commit hooks can check, if the
published information conforms to predefined requirements.
Furthermore, we want to investigate, which data quality metrics can be applied
on research data with Linked Data annotations. Additional data quality metrics
for research data may be defined and implemented. This especially applies for
metrics that ensure scientific integrity, conformance between the research data
itself and the claimed meta information and good scientific practice.
Finally, data quality checks can not only be run in the publication process but
for open research data also on data sources in a remote fashion. In combination
with the research data discovery and query, we may extend our preliminary work
on the definition of data quality assessment requirements (DaQAR).
Fig. 3. Basic conceptual architecture
49
5.4 Research methodology
The PIROL project will focus on a mixed qualitative and quantitative research
in information management. It will yield both a concept for systematic and
sustainable cross-domain data annotation and publishing activities of research
data as well as tested techniques on how to ensure appropriate and gainful
cross-domain research data provision from a human factor perspective.
For project preparation purposes, a survey among multiple researchers from
different knowledge domains will be conducted to get a basic understanding
on the characteristics, content and amount of available research data. New
techniques and extensions will then be designed, implemented and tested in an
incremental, iterative fashion in small developer teams, that have expertise in Web
Engineering, Knowledge Management and Component-oriented Development to
develop solutions that are reusable in different contexts. Developed strategies and
tools will be evaluated with continuous user questionnaires and lab experiments
to which extend they satisfy pre-defined requirements.
6 Contribution
The PIROL research project will contribute a research data publishing process
with independent components as depicted in fig. 3 to the problem domain of
interdisciplinary and decentralized research data management:
Meta data profile for research data annotation defined by comparing
and homogenizing existing vocabularies for interdisciplinary usage
Registry for concept URIs in Linked Research Data established To
the best of our knowledge,no similar service exists so far
Research data annotation tool provided To the best of our knowledge, no
established platform-independent tool exists so far
Decentralized research data management with SoLiD approved To
the best of our knowledge, no Proof-of-concept has been done so far
SoLiD extenders for common data repositories implemented by
providing a concept on how to extend traditional data platforms
Query mechanism on SoLiD based research data stores evaluated by
applying and extending common query strategies on research data stores
Content of data stores communicated to data catalogs by improving
mechanisms to harvest and update (meta)data records with Linked Data
Particular data quality metrics for research data defined by providing
algorithms on how to automatically assess relevant aspects
Automated data conformance and quality assessment implemented
as data is published without automated check operations so far
Ontology to describe data requirements and characteristics specified
by extending existing approaches such as DaQAR and SHACL on how to
describe research data characteristics of interest
50
7 Conclusion
The digital transformation led also in research to new challenges in data ma-
nagement, which includes interdisciplinary annotation and discovery as well as
data-privacy issues. The objective of the presented PIROL: Publishing Interdisci-
plinary Research On Linked data is to improve the publishing process of existing
(thus persistently stored) referenceable research data for interdisciplinary search
and reuse with the means Linked Data. This includes a separate annotation step
with a human-centered user input interface concept to create a meta descrip-
tion for concrete research data, the validation of data conformance and data
quality aspects between this meta profile and the original research data, and
the publishing in an LDP enabled, decentralized infrastructure.
References
1. Bizer, C., Heath, T., Berners-Lee, T.: Linked Data - The Story So Far.
IJSWIS 5(3), 1–22 (jul 2009)
2. Griffin, P.C., Khadake, J., LeMay, K.S., et al.: Best practice data life cycle
approaches for the life sciences. F1000Research 6, 1618 (jun 2018)
3. Khalili, A., Loizou, A., van Harmelen, F.: Adaptive linked data-driven web
components: Building flexible and reusable semantic web interfaces. Lecture
Notes in Computer Science 9678, 677–692 (2016)
4. Langer, A., Gaedke, M.: Fame.Q - A formal approach to master quality in
enterprise linked data. In: Proceedings of the 15th International Conference
WWW/Internet 2016. No. October (2016)
5. Langer, A., Gaedke, M.: DaQAR - An ontology for the uniform exchange
of comparable LD quality assessment requirements. In: Lecture Notes in
Computer Science. vol. 10845 LNCS, pp. 234–242. Springer, Cham (jun 2018)
6. Langer, A., Göpfert, C., Gaedke, M.: URI-aware user input interfaces for
the unobtrusive reference to Linked Data. IADIS International Journal on
Computer Science and Information Systems 13(2) (2018)
7. Langer, A., Siegert, V., Göpfert, C., Gaedke, M.: SemQuire - assessing the
data quality of linked open data sources based on DQV (2018)
8. Mansour, E., Sambra, A.V., Hawke, S., et al.: A Demonstration of the Solid
Platform for Social Web Applications. Proceedings of the 25th International
Conference Companion on World Wide Web pp. 223–226 (2016)
9. Sousa, R.B., Cugler, D.C., Malaverri, J.E.G., Medeiros, C.B.: A provenance-
based approach to manage long term preservation of scientific data. In: 2014
IEEE 30th ICDE Workshops. pp. 162–133. IEEE (mar 2014)
10. Steinhof, C.: Erfolgskriterien von Forschungsdatenrepositorien und deren
Relevanz für verschiedene Stakeholder-Gruppen (2017)
11. Umbrich, J., Hogan, A., Polleres, A., Decker, S.: IOS Press Link Traversal
Querying for a Diverse Web of Data. Semantic Web Interoperability, Usability,
Applicability 0 (2014)
12. Zaveri, A., Rula, A., Maurino, A., et al.: Quality Assessment for Linked Open
Data: A Survey. Semantic Web Journal (by IOS Press) 1, 1–31 (2014)
51