Technical ReportPDF Available

Data Ecosystems: Sovereign Data Exchange among Organizations. Report from Dagstuhl Seminar 19391


Abstract and Figures

This report documents the program and the outcomes of Dagstuhl Seminar 19391 "Data Ecosystems: Sovereign Data Exchange among Organizations". The goal of the seminar was to bring together people from different disciplines (also outside the computer science area), in order to identify (i) a set of research challenges for the future development of data ecosystems and a catalogue of major approaches relevant to the field and (ii) a set of developed use cases of particular interest to the further development of data ecosystems. Towards the objectives, the seminar included tutorials, invited talks, presentations of open problems, working groups. This report presents the most relevant findings and contributions. Seminar September 22-27, 2019-
Content may be subject to copyright.
Report from Dagstuhl Seminar 19391
Data Ecosystems: Sovereign Data Exchange among
Edited by
Cinzia Cappiello1, Avigdor Gal2, Matthias Jarke3, and
Jakob Rehof4
1 Polytechnic University of Milan, IT,
2 Technion – Israel Institute of Technology – Haifa, IL,
3 RWTH Aachen, DE,
4 TU Dortmund, DE,
This report documents the program and the outcomes of Dagstuhl Seminar 19391 “Data Ecosys-
tems: Sovereign Data Exchange among Organizations”. The goal of the seminar was to bring
together people from different disciplines (also outside the computer science area), in order to
identify (i) a set of research challenges for the future development of data ecosystems and a cata-
logue of major approaches relevant to the field and (ii) a set of developed use cases of particular
interest to the further development of data ecosystems. Towards the objectives, the seminar
included tutorials, invited talks, presentations of open problems, working groups. This report
presents the most relevant findings and contributions.
Seminar September 22–27, 2019 –
2012 ACM Subject Classification Information systems Information integration, Information
systems Data exchange, Information systems Ontologies, Computing methodologies
Machine learning, Computer systems organization Architectures, Security and Privacy,
Information systems Data analytics, Software and its engineering, Social and professional
topics Computing and business
Keywords and phrases Data sovereignty, Data ecosystems, Business models, Data integration,
Digital Object Identifier 10.4230/DagRep.9.9.66
Edited in cooperation with Bernadette Farias Lóscio (Collector)
1 Executive Summary
Cinzia Cappiello (Politecnico di Milano, IT)
Avigdor Gal (Technion – Haifa, IL)
Matthias Jarke (RWTH Aachen, DE)
Jakob Rehof (TU Dortmund, DE)
License Creative Commons BY 3.0 Unported license
©Cinzia Cappiello, Avigdor Gal, Matthias Jarke and Jakob Rehof
The design of data ecosystems, infrastructures for the secure and reliable data exchange among
organizations, is considered as one of the key technological enablers for digitalization and
the digital economy of the future. Several applied research initiatives and industry consortia
provide substantive evidence of this trend e.g., the Industrial Internet Consortium (IIC)
Except where otherwise noted, content of this report is licensed
under a Creative Commons BY 3.0 Unported license
Data Ecosystems: Sovereign Data Exchange among Organizations, Dagstuhl Reports, Vol. 9, Issue 9, pp. 66–134
Editors: Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof
Dagstuhl Reports
Schloss Dagstuhl – Leibniz-Zentrum für Informatik, Dagstuhl Publishing, Germany
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 67
formed in the USA, the Industrial Data Space (IDS) founded in Germany and the associated
consortium International Data Space Association (IDSA)
. Most of these initiatives aim
to provide a reference architecture for dealing with (i) governance aspects related to the
definition of policies and conditions able to norm the participation to the data ecosystem,
(ii) security aspects related to the definition of policies and infrastructures for guaranteeing
a trusted and secure exchange of data, (iii) data and service management aspects related
to representation models and exchange formats and protocols, and (iv) software design
principles related to the realization of the architectural components and their interaction.
All these aspects have been discussed in the seminar and the main findings are described
in this report. In addition, a central new aspect of data ecosystems that we considered in
the seminar lies in the view of data as having an economic value next to its intrinsic value
to support operational and decisional core business activities. This means that in the data
ecosystem, data is typically considered both a business asset and a business commodity
which may be priced and sold in some form (e.g., data provisioning service or raw data)
according to contracts.
As testified by the amount and variety of problems described above, the creation of such
ecosystems poses many challenges cutting across a wide range of technological and scientific
specializations. For this reason, the seminar involved researchers from different communities.
Interdisciplinary discussions gave the possibility to analyze different perspectives and to
achieve valuable outcomes presented in this report, such as a wide set of research challenges
and the definition of interesting use cases for the further development of data ecosystems.
Details about the activities carried out during the seminar are provided in the following.
Overview of the activities
The seminar took place from Monday September 23 until Friday September 27. The seminar
program encompassed four invited talks (keynotes and tutorials) on the first day (Sep. 23rd),
by Gerald Spindler (law and ethics), Frank Piller (ecosystems and business models), Maurizio
Lenzerini (data integration), and Boris Otto (International Data Space). After discussions
related to the talks and tutorials, the remaining afternoon was spent structuring (through
joint discussion) the coming days of the seminar and group structure. As a result, group
structure was based on a thematic structure encompassing three groups, one for each of
the topic areas Business, Data, and Systems. Tuesday Sept. 24 began with a breakout into
groups and election of scribes in each of the three groups (Business, Data, and Systems),
and the remainder of the day was taken up by parallel group sessions in the three groups.
Wednesday Sept. 25 began with a joint session where each of the groups presented their
work, which was then discussed jointly. The afternoon (until the excursion) was taken up
by joint discussion on report structure. The morning of Thursday Sept. 26 encompassed
joint discussion on a proposed joint manifesto as well as group discussions on application
domains and application scenarios (topic areas were Health, SmartCities, Industry 4.0). The
afternoon was taken up by continued group discussions and ended with group presentations
and joint discussion on application domains and application scenarios. There was also further
discussion on report structure at the end of the day. The manifesto was subject to very lively
discussion in the evening, after dinner. Friday Sept. 27, the last day of the seminar, was
68 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
devoted to wrap-up (conclusions, summary, and report process) followed by joint discussion
on relations between Systems, Data and Business views on the overall topic of the seminar.
The outcome of the seminar, which is documented in the remainder of this report,
encompasses summaries of the group discussions and the joint manifesto.
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 69
2 Table of Contents
Executive Summary
Cinzia Cappiello, Avigdor Gal, Matthias Jarke and Jakob Rehof ........... 66
Overview of Invited Plenary Talks
Recent developments of a legal framework for IT
Gerald Spindler ...................................... 71
A few thoughts about managing business models for platform-based data ecosystems
Frank Piller ........................................ 71
Semantic Data Interoperability
Maurizio Lenzerini .................................... 71
International Data Spaces
Boris Otto ......................................... 72
Boris Düdder, Wolfgang Maaß, Julian Schütte ..................... 72
Cinzia Cappiello, Yuri Demchenko, Ugo de’Liguoro, Bernadette Farias Loscio,
Avigdor Gal, Sandra Geisler, Maurizio Lenzerini, Paolo Missier, Elda Paja, Barbara
Pernici, Jacob Rechof, Simon Scerri, Maria-Esther Vidal .............. 77
The Business of Data Ecosystems
Elda Paja, Matthias Jarke, Boris Otto, Frank Piller ................. 85
Use Cases
Use Cases from the Medical Domain
Sandra Geisler, Maria-Esther Vidal, Elda Paja, Maurizio Lenzerini, Paolo Missier 95
Industrie 4.0 Data Ecosystems Examples
Egbert-Jan Sol ......................................102
Use Cases from the Smart Cities Domain
Cinzia Cappiello, Bernadette Farias Lóscio, Avigdor Gal, Fritz Henglein ......108
Manifesto ..........................................110
Statements of the participants .............................111
Cinzia Cappiello (Polytechnic University of Milan, IT) . . . . . . . . . . . . . . . . 111
Ugo de’Liguoro (University of Turin, IT) . . . . . . . . . . . . . . . . . . . . . . . 111
Yuri Demchenko (University of Amsterdam, NL) . . . . . . . . . . . . . . . . . . . 111
Elena Demidova (Leibniz Universität Hannover, DE) . . . . . . . . . . . . . . . . . 112
Boris Düdder (University of Copenhagen, DK) . . . . . . . . . . . . . . . . . . . . 113
Bernadette Farias Lóscio (Federal University of Pernambuco, BR) . . . . . . . . . 113
Avigdor Gal (Technion Haifa, IL) . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Sandra Geisler (Fraunhofer FIT Sankt Augustin, DE) . . . . . . . . . . . . . . . 114
70 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
Benjamin Heitmann (Fraunhofer FIT – Aachen, DE & RWTH Aachen, DE) . . . . 115
Fritz Henglein (Univ. of Copenhagen, DK & Deon Digital – Zürich, CH) . . . . . . 115
Matthias Jarke (RWTH Aachen University and Fraunhofer FIT, DE) . . . . . . . . 116
Jan Jürjens (Universität Koblenz-Landau, DE) . . . . . . . . . . . . . . . . . . . . 117
Maurizio Lenzerini (Sapienza University of Rome, IT) . . . . . . . . . . . . . . . . 117
Wolfgang Maaß (Universität des Saarlandes Saarbrücken, DE) . . . . . . . . . . 119
Paolo Missier (Newcastle University, GB) . . . . . . . . . . . . . . . . . . . . . . . 119
Boris Otto (Fraunhofer ISST – Dortmund, DE & TU Dortmund, DE) . . . . . . . 119
Elda Paja (IT University of Copenhagen, DK) . . . . . . . . . . . . . . . . . . . . 120
Barbara Pernici (Polytechnic University of Milan, IT) . . . . . . . . . . . . . . . . 121
Frank Piller (RWTH Aachen, DE) . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
Andreas Rausch (TU Clausthal, DE) . . . . . . . . . . . . . . . . . . . . . . . . . . 122
Jakob Rehof (TU Dortmund, DE) . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Simon Scerri (Fraunhofer IAIS Sankt Augustin, DE) . . . . . . . . . . . . . . . . 124
Julian Schütte (Fraunhofer AISEC München, DE) . . . . . . . . . . . . . . . . . 125
Egbert Jan Sol (TNO Eindhoven, NL) . . . . . . . . . . . . . . . . . . . . . . . . 126
Maria-Esther Vidal (TIB Hannover, DE) . . . . . . . . . . . . . . . . . . . . . . 126
Participants .........................................134
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 71
3 Overview of Invited Plenary Talks
3.1 Recent developments of a legal framework for IT
Gerald Spindler (University of Göttingen, DE)
License Creative Commons BY 3.0 Unported license
©Gerald Spindler
In his highly stimulating and provocative keynote address, law professor and high-level
EU advisor Gerald Spindler shared important observations about a serious misfit between
the conceptualizations pursued in business administration and computer science, and the
structuring of the law system. As a consequence, judges are often surprised by seemingly
unpredictable and contradictory answers to their legal questions. Conversely, managers and
engineers are confronted with a law system that seems to adapt extremely slowly to the
rapid progress, and with for them very surprising interpretation of this changing reality. For
example, the speaker surprised the audience with the statement, that no concept of data
ownership exists in Europe, except for the right to ones own personal data in the GDPR
regulation. In the discussion, all participants agreed that joint research is urgently needed to
better match the conceptual world of law and ethics, with the technical, user, and business
perspectives on data ecosystems.
3.2 A few thoughts about managing business models for
platform-based data ecosystems
Frank Piller (RWTH Aachen University, DE &MIT Media Lab, US)
License Creative Commons BY 3.0 Unported license
©Frank Piller
In his keynote, Frank Piller started from the observation that – in contrast to traditional
product platforms e.g., in the automotive industry – the value of smart products does no
longer lie in the product itself, but rather in the connections in its business ecosystem.
The resulting network effects and multi-sided markets have already been intensely studied,
most visibly by 2015 Economics Nobel Prize winner Jean Tirole. The strategic question
then becomes whether a company wants to offer a platform or an app. Successful platform
businesses show striking differences from traditional organizations e.g., in terms of value
created vs. value captured per employee. Current empirical research at RWTH Aachen
University is studying how these concepts are being transferred to networks of German small
and medium enterprises, e.g., in the smart farming sector.
3.3 Semantic Data Interoperability
Maurizio Lenzerini (Sapienza University of Rome, IT)
License Creative Commons BY 3.0 Unported license
©Maurizio Lenzerini
In his tutorial, Maurizio Lenzerini gave an logic-based structuring of four different data
interoperability architectures: data integration, data exchange, data warehouses/data lakes,
and collaborative data sharing. He pointed out the central importance of formal mappings
72 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
with different schemas, and then went into some depth into description logic-based approaches
for compile time tasks such as mapping discovery, analysis, and reasoning, but also into
runtime tasks such as data exchange, update propagation, data quality, and mapping-based
direct and reverse query rewriting.
3.4 International Data Spaces
Boris Otto (Fraunhofer ISST &TU Dortmund, DE)
License Creative Commons BY 3.0 Unported license
©Boris Otto
In his overview talk on the Fraunhofer-led International Data Spaces (IDS) initiative, Boris
Otto first presented some typical examples from European industries, and cited worries
of a majority of European SMEs about trust and security, data sovereignty in terms of
keeping control over shared data, and inconsistencies in not just data interoperability, but
also process interoperability. He then reported a large-scale, multi-year requirements study
about desirable properties of a solution, from which the IDS reference architecture as well as
a conceptual information model, several advanced algorithms (e.g., for data usage control)
and suitable governance mechanisms for so-called alliance-driven platforms are emerging.
4 Technology
4.1 Systems
Boris Düdder (Univerity of Copenhagen, DK) and Wolfgang Maaß (Saarland University,
DFKI, DE) and Julian Schütte (Fraunhofer AISEC, DE)
License Creative Commons BY 3.0 Unported license
©Boris Düdder, Wolfgang Maaß, Julian Schütte
4.1.1 Motivation
Data ecosystems operating valuable data assets need strong guarantees on data security
while fostering openness, community, and value creation.
Data is becoming a significant asset for any organization [
]. A key challenge is related to
technical architectures for managing and processing data. A dominant approach centralizes
data and distributes processing results (ref). Alternatives are less popular but feasible, such
as peer-to-peer architectures or other more distributed approaches (ref) in which either
data or program (query) are transferred optimizing for connection capacities. Related to
the underlying architecture are organizational principles and governance structures. For
centralized architectures, principal-agent logic is applied, i.e., a principal (user, client,
customer) sends data to an agent (server, service provider) who processes the data and
sends back results. This organizational architecture is applicable if the key value lies in the
software as such, but not in the data. Now that data become a value as such, a bilateral
principal-agent, or technically a client-service architecture, does not necessarily fulfill the
requirements of data economic systems.
Several technical elements are essential key components for a data ecosystem, as explained
in the following subsections, such as security and encryption, data analytics, AI and semantic
technologies and ontologies, multi-agent systems, peer-to-peer architectures and, last but not
least, algorithmic correctness and correctness of implementations.
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 73
4.1.2 Scope and Requirements
The system boundaries are used to define the purpose of our data ecological system w.r.t. tech-
nologies and applications. Therefore, these system boundaries influence fundamental design
decisions of the problem model design and are reflected in the solution design. The boundaries
are defined along the dimensions of the organization and technical scope. Organizational
boundaries are organized along with entities’ roles, rights, obligations, and prohibitions. Such
entities are key concepts such as contracts specifying obligations, data/data objects/data sets
representing valuable curated data assets, and semantics/metadata/ontologies generating
information out of pure data. The technical scope should support system design and modules,
various pervasive architectures, i.e., central, tightly coupled, and loosely coupled, needs for
adaptability [12] and security, as well as protocols, computation and programming models.
Data formats and protocols are necessary for expressing contracts, i.e., for enforcing
policies for data usage and conditions or guards such as pricing or billing for example. A
challenge in a distributed network to enforce contracts and usage constraints at a remote peer,
which is controlled by another participant. Possible ways for enforcement are by establishing
trust in the participants and their data handling, e.g., employing remote attestation [
], or
by private function evaluations [21].
4.1.3 State of the Art
Traditionally data is not perceived as a valuable asset as such but as an intrinsic part of
any software. Data populates databases as a carrier of facts about a domain. Physicists,
chemists, astronomers, biomedicial researchers, and economists understand for a long time
that data is a basic asset that requires processing and filtering for deriving knowledge (e.g.,
CERN LHC). Companies, such as Google, Amazon, Microsoft, Baidu, and Alibaba, adopted
this understanding for extracting knowledge that can be used for predicting human behavior.
The success of these attempts was transferred to production industries (Industrie 4.0 and
cyber-physical systems) and even private life (smart city and smart home). Thus, data
becomes a valuable asset as such, called a data product [
]. Taking a product-driven
approach, system designs start with the data product and ask in which market, to whom,
and for what price it can be sold, i.e., applying a product logic. This adopts an inside-out
perspective by which a data product is the starting point, and the market is the target. From
an inside-out perspective, prices for data products are determined by adopting a cost-driven
approach. Taking an outside-in perspective, the customer of a data product is the starting
point by asking the question of which value can be created on the customer side by leveraging
data products. This generally applies a service logic.
Data ecosystems are characterized by the type of business and the type of community [
The terms ‘knowledge market’ and ‘data market’ are used as synonyms nowadays. The
type of business is distinguished between commercial and non-commercial data ecosystems
while the type of community distinguishes between closed on open. Closed data ecosystems
are established under the umbrella of one juridical person. This could be, for instance, a
corporation, an association, or an individual person. Non-commercial, closed data ecosystems
are intraorganizational while non-commercial, open data ecosystems provide data products
unlimited for free or with the option for donations, e.g., UC Irvine Machine Learning Repos-
itory. Viable business models for data ecosystems are closed, for instance, by membership or
proprietary bi- or multi-lateral contracts. Commercial, open data ecosystems are organized
as marketplace [24] that support trading and transactions [4].
74 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
Figure 1 Basic system design patterns for data ecosystems.
System designs of data ecosystems taking a product logic adopt the concept of a mar-
ketplace in which the main interaction occurs by matching data product providers with
customers. Facilitating roles are providing support services on financial transactions, logistics,
quality assurance, insurance, and notary services. The main purpose of a product-driven
marketplace lies in the transfer of access and usage rights under governance of explicit
contracts or background regulations and laws.
The most simple system design is a data
pipeline that establishes a marketplace between a single data provider and a single customer.
Facilitating services are fully materialized by the marketplace and fixed contracts. More
flexible are data hubs that allow
transactions between data providers and customers.
: 1 data hub is used if there is only one customer but many data providers. Dominant
customers establish data hubs and invite data providers (dominant customer data hub). In
contrast, a 1 :
data hub has only one or very few data providers but many customers (e.g.,
Airbus Skywise or Bloomberg Market Data Feed). A 1 :
data hub system design implements
a standard product distribution network (dominant supplier data hub). A derivative of this
system design is a 1:1:n data hub with an independent agent implementing a data hub
between one data provider and
customers. In reverse, the same system design can be
applied with one dominant customer, i.e., n: 1 : 1 data hub.
More elaborated are
data hubs or
: 1 :
data hubs that constitute the concept of
adata marketplace. Without dominant market players, a data marketplace is governed by
offer and demand [
]. A data marketplace is considered being liquid if any demand can be
contractually satisfied by an offer for a mutually acceptable price and low transaction costs.
In plain data marketplace, trust between buyers and sellers is not guaranteed but requires
additional trust-building regulations and additional roles.
The support for trusted and secure computation balancing participants requirements and
expectations is challenging. Electronic marketplaces with multiple agents have been studied
under various security objectives, including privacy, non-refutation, and accountability. The
focus has primarily been on the digital transactions of such markets and how to secure them.
The physical value-chain is aligned to the digital transactions of such marketplaces because
According to Gerald Spindler’s statements, data ownership cannot be transferred from a data originator
to any other agent.
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 75
Figure 2 Generic system design for data marketplaces.
these marketplaces emphasize on physical goods as transaction value and not on data as a
valuable good. To secure consistency and correctness of transactions, at least two different
approaches are in use. The first approach employs monitoring and auditing of the processes,
checking the consistency of good transfers and digital representation, as well as ensuring that
data flows adhere to contractual agreements. The costly and laborious approach motivated
due diligence [
] and various usage control frameworks [
]. Another approach is
prevention by either using legal penalties to force compliant behavior of market participants
or by technical measures including cybersecurity and cryptography. The latter has become a
highly active area of research and several significant advances have been made in privacy-
enhancing technologies such as verifiable computation and non-interactive zero-knowledge
proofs [
], distributed ledger technology [
], and fully-homomorphic
cryptography [
]. Depending on the desired guarantees for the system, the correctness
and robustness of algorithms and the system implementations are necessary. Security is not
a product; it is a process, which demands a verified tool-chain, e.g., akin CompCert [
and platform, i.e., trusted execution environments. The interaction of components and their
reliability could be supported by automatic program generation [
], DSLs [
] and
verification methods.
4.1.4 Model
Data pipelines and data hubs replicate traditional interactions in business relationships
between firms. Data marketplaces are a model class subsuming all other types of data
ecosystems. Three roles are key to a data marketplace: buyer, seller, intermediary trusted
agent. The role of a trusted agent is mandatory because of the transactional nature of a
marketplace that requires a transaction being atomic. Data products
are provided from
to buyer
in exchange to another entity, in particular, financial assets. By doing
this, value is created on the buyer side. In general, a seller
only transfers projection of
to buyer
for various reasons, such as IP protection and trade secrets. Nonetheless,
a seller
might have a business incentive for selling certain properties and insights on
Therefore, seller
and buyer
negotiate a function
that shall be allowed to apply on
under contractual requirements
. The role of the trusted agent
is that only
is applied
to Xas specified in C(Figure 2).
X, C
A, B
)) :
means that a function
is applied to a data product
or a set of data products according to specification given by contract
acting as a callback
function in
, e.g., in [
]. The result is provided to buyer B according to specifications in
. Function
is required to be implemented in a stateless fashion, and no traces
will be left with agent T. These needs to be assured by trust-building measures, such as
program verification and certification. Therefore, a system design consisting of seller A,
76 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
buyer B, and a trusted agent T guarantees that a) data products are only used according to
a specification, b) buyer B only gains access to resulting data, and c) no data traces are left
with agent T.
4.1.5 High-level Description of Possible Applications
We present here two use case scenarios which could benefit from such a system.
Predictive maintenance – Predictive maintenance helps to determine the condition of
operational equipment in-use for predicting when maintenance should be performed and
which components are affected. Cost savings over routine or time-based preventive main-
tenance could be achieved because activities are triggered only when necessary. In this
scenario, a production site creates raw data from production machine sensors, and the
machine manufacturer is interested in running analytics on the sensor data to improve his
customer maintenance service. On the other hand, the production plant owner is concerned
about sharing raw data of the production machines, which could allow interested parties,
e.g., competitors, to infer business secrets. The scenario can be enabled with a secure data
sharing which balances the interests of both parties or even third parties.
Pharmaceutical companies – Pharmaceutical companies are producing and delivering
drugs to patients through complex supply chains. The associated indirection between
pharmaceutical company and patient complicates the due diligence, e.g., in the presence
of adverse effects of drugs. Thus, sharing information between pharmaceutical companies,
regulators, health practitioners, and patients could improve drug safety and reduce the costs
of due diligence. All parties in this scenario have interests to share data but, at the same
time, have high interests in restricting the data exchange, e.g., for privacy or competitive
advantages. The system is realizable using secure data sharing allowing to create a data
ecosystem harmonizing the needs of all participants.
4.1.6 Research Challenges
The project is ambitious, and to our best knowledge, there is no product on the market
dealing with the identified challenges on business, legal, and computer science side. Business
is founded on trust and trust-building activities, which is expensive to build and to maintain.
The research challenges identified are (semi-)automatic contracts negotiation and agreement
on functions as well as function execution. Semantic information of data and functions is
necessary to automate negotiation and agreement on functions. On the other hand, data-
/function-specific guarantees of market participants need to comply with data protection
rules. The treatment of data in law is very often focused on privacy but not on data as an
asset. The consequent juridical uncertainty, e.g., on the data ownership and copyright, is
a business risk and poses a challenge for law researchers as well as legislation. To replace
trust with mathematical and technological guarantees is an additional research challenge
for computer science as a discipline. The verification of functions, their execution in secure
execution environments, and secure and trusted libraries are not available on the market
and require research from various interdisciplinary and inter-subdisciplinary research. The
system challenges are compelling but indispensable for the future success of data ecosystems.
4.1.7 Conclusion
Because of the interdisciplinary aspects of data ecosystem platforms, computer science can
provide various models, methods, and tools to support the development of such platforms.
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 77
The challenges are not only located in computer science but also the interplay with other
disciplines. The challenges and opportunities presented in this section are novel in combination
with their usage domain and its inherent restrictions, e.g., legal or economic. We postulate
a platform and sound formal models supporting the activities described in this report and
offering solutions to the more general problems of data ecosystems.
4.2 Data
Cinzia Cappiello (Polytechnic University of Milan), Yuri Demchenko (University of Am-
sterdam, NL), Ugo de’Liguoro (University of Turin, IT), Bernadette Farias Lóscio (Fed-
eral University of Pernambuco, BR), Avigdor Gal (Technion – Haifa, IL), Sandra Geisler
(Fraunhofer FIT – Sankt Augustin, DE), Maurizio Lenzerini(Sapienza University of Rome,
IT), Paolo Missier (Newcastle University, GB), Barbara Pernici (Polytechnic University
of Milan, IT), Jacob Rehof (TU Dortmund, DE), Simon Scerri (Fraunhofer IAIS – Sankt
Augustin, DE), Maria-Esther Vidal (TIB – Hannover, DE)
License Creative Commons BY 3.0 Unported license
©Cinzia Cappiello, Yuri Demchenko, Ugo de’Liguoro, Bernadette Farias Loscio, Avigdor Gal,
Sandra Geisler, Maurizio Lenzerini, Paolo Missier, Elda Paja, Barbara Pernici, Jacob Rechof,
Simon Scerri, Maria-Esther Vidal
4.2.1 Scope
In this section, the main outcomes of the group of Data are reported. First, diverse data-
driven problems are presented; solutions to these problems demand to effectively describe
and integrate heterogeneous data, accurately represent and assess data quality, and ensure
data specifications during the execution of operators or queries. Next, the state of the art is
summarized and the proposed model for data ecosystems is defined. Finally, insights of the
research challenges to be addressed are outlined.
4.2.2 Motivation and Requirements
Data-driven technologies in conjunction with smart infrastructures for management and
analytics, increasingly offer huge opportunities for improving quality of life and industrial
competitiveness. However, data has grown exponentially in the last decades as a result
of the advances in the technologies for data generation and ingestion. Moreover, data is
usually ingested in myriad unstructured formats and may suffer reduced quality due to
biases, ambiguities, and noise. Thus, the development of efficient data management methods
are demanded for enabling the transformation of disparate data into knowledge from which
actions can be taken in scientific and industrial domains. Problems to be addressed include:
Definition of a generic data market architecture that can describe data properties such
as economic goods and data exchange models. In this type of architectures, data lakes
play a relevant role in the storage of huge amount of heterogeneous data on distributed
Active data networks described in terms of specifications and collections of components
which can be repositories of data or queries. Data sources are connected using schema
mappings. The evaluation of queries against the network requires the composition of data
sources respecting the restrictions imposed by the connections existing the components
of the network.
78 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
Accurate description of meta-data of data sources that can change over time and effective
usage of these descriptions whenever data is shared.
Models that estimate the cost of integrating different sources, and the benefits that the
fusion of new data sources adds to the accuracy of query processing.
Hybrid approaches that combine computational methods with human knowledge; limit-
ations and benefits of these approaches in the resolution of data-driven problems, e.g.,
schema matching, data curation, and data integration need to be established.
Paradigms for making data suitable for sharing are required; they should ensure trust-
worthiness, e.g., in terms of data quality or data market players.
Federated query processing for addressing data interoperability issues during query
execution demands meta-data to select the relevant sources for a query. Furthermore,
data quality assessment is required for determining the quality of the answers produced
during query execution against a set of selected sources.
Management of new generated data from data analysis and machine learning performed
on existing sources.
4.2.3 State of the Art
4.2.4 Data Integration
The problem of integrating data collected from different data sources has been extensively
treated in the literature [
]. The mediator and wrapper architecture proposed by
Wiederhold [
] and the data integration system approach presented by Lenzerini [
represent the basis for the state of the art [
]. Following these approaches a
global ontology encodes domain knowledge and enables the description of the meaning of
the data to be integrated by means of mapping rules. Since creating mappings manually in
a data integrating system is a tedious and time consuming task, diverse approaches have
been proposed to discover mappings in a (semi-)automated way (e.g., CLIO[
], IncMap [
GeRoMe [
], KARMA [
]). Additionally, a vast amount of research has been conducted
to propose effective and efficient approaches for ontology alignment or schema matches [
]; modeling and management of uncertainty of the schema matching process
is the paramount importance for providing a quantifiable analysis of the accuracy of the
discovered mappings [44].
4.2.5 Data Ecosystems
Data ecosystems are a special kind of digital ecosystem. As such they are distributed, open
and adaptive systems with the characteristics of being self-organizing, scalable and sustainable.
While being centered on data, the main concern with data ecosystems is about knowledge
sharing and growing, which is at the same time an issue of learning from unstructured
and heterogeneous data, construction of new abstractions and mappings, offer of services,
including querying, data integration and transformation. All this should be ensured in a
dynamic and scalable way, while retaining consistency, quality assessment, security and
4.2.6 Query Processing Over Heterogeneous Data Sets
Existing solutions to the problem of query processing over heterogeneous data sets rely
on a unified interface for overcoming interoperability issues, usually based on metamodels
]. A few Data Lake systems have been proposed, mainly with focus on data ingestion
and metadata extraction and management. Exemplary approaches include GEMMS [
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 79
Data Sets
Data Operators
Descriptions Properties Domain Ontologies
(a) A Basic Model for a Data Ecosystem.
(b) A Data Ecosystem equipped with a brain.
Figure 3 A Basic Model for a Data Ecosystem.
PolyWeb [
], BigDAWG [
], Ontario [
], and Constance [
]. These systems collect
meta-data about the main characteristics of the heterogeneous data sets in a Data Lake,
e.g., formats and query capabilities; additionally, they resort to a global ontology to describe
contextual information and relationships among data sets. Rich descriptions of the properties
and capabilities of the data have shown to be crucial for enabling these systems to effectively
perform query processing.
4.2.7 A Model for Data Ecosystems
A data ecosystem DE is defined as a 4-triple DE=<Data Sets, Data Operators, Meta-Data,
Mappings>; Figure 3a depicts a data ecosystem in terms of its components.
Data sets: the ecosystem is composed of a set of data sets. Data sets can be structured
or unstructured; also, they have different formats, e.g., CSV, JSON or tabular relations,
and can be managed using different management systems.
Data Operators: the set of operators that can be executed against the data sets.
Meta-Data: provides the description of domain of knowledge, i.e., the meaning of the
data stored in the data sets of the data ecosystem. It comprises:
Domain ontology provides a unified view of the concepts, relationships, and constraints
of the domain of knowledge. It associates formal elements from the domain ontology
to each
. For instance, workshop and participant can be part of the concepts in a
domain ontology.
Properties enable the definition of data quality, provenance, and data access regula-
tions of the data in the ecosystem. For instance, last updated and other non-domain
properties (quality etc).
Descriptions of the main characteristics of a data set. No specific formal language or
vocabulary is required; in fact, a data set could be described using natural language.
For instance, Data set D is about a Dagstuhl seminar.
Mappings expressing correspondences among the different components of a data ecosystem.
The mappings are as follows:
80 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
Data Sets
Data Operators
Descriptions Properties Domain Ontologies
Data Sets
Data Operators
Descriptions Properties Domain Ontologies
Descriptions Properties Domain Ontologies
Node 1 Node 2
D1 D2
Figure 4 A Model for a Network of Data Ecosystems.
Mappings between ontologies: they represent associations between the concepts in the
different ontologies that compose the domain ontology of the ecosystem.
Mappings between the data sets: they represent relations among the data in the data
sets of the ecosystem and the domain ontology.
A data ecosystem can be equipped with a “brain”, able to execute services against the data
sets (Figure 3b). Services include query processing, data transformation, anonymization, data
quality assessment, or mapping generation. The services are able to exploit the knowledge
encoded in the meta-data and operators to satisfy the requirements of the applications
implemented across the ecosystem. The following correspond to examples of services:
Concept discovery: identify a new concept, e.g., foreign student, using machine learning.
Based on the result, the domain ontology and the mappings can be augmented.
Data set curation: a service able to keep humans in the loop can be used to create a
curated version of a data set in the ecosystem. The service can also update the properties
of the ecosystem to indicate the provenance of the new curated data set and manage new
generated data from data transformation, analysis, and learning.
Procedure synthesis: a service able to construct new procedures out of elementary build
blocks by composing existing services toward new goals. In a complex and evolving
system, it would be unfeasible to program procedures and even queries without automatic
support; also, the exploration of repositories and libraries of existing procedures should
be available.
A set of data ecosystems can be connected in a network. For this, we envision an
ecosystem-wide meta-data layer where the entire ecosystem is described. Figure 4 depicts a
network where nodes and edges correspond to data ecosystems and mappings among them,
respectively. In this configuration, the meta-data layer describes each of the nodes in terms
of descriptions, properties, and domain ontologies. The following types of mappings can be
defined among the nodes of a network of data ecosystems:
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 81
Data Sources
Data Operators
Descriptions Properties Domain Ontologies
Data Sources
Data Operators
Descriptions Properties Domain Ontologies
Descriptions Properties Domain Ontologies
Strategy/Business Strategy/Business
Stakeholders Objectives dependencies
Node 1
Data exchange
Node N
Trust Agreements
Figure 5
A Model for a Network of Data Ecosystems Empowered with Strategy and Business
Models and Regulations.
Mappings between domain ontologies: they state correspondences among the domain
ontologies of two nodes or between one node and the global meta-data layer. For example,
the concepts workshop and seminar in nodes N1and N2, respectively, are the same.
Mappings between properties: they describe relationships among properties in two nodes.
For example, the provenance of two curated versions of a data set could be the same.
Mappings between data sets: they represent correspondences between the data sets in two
nodes. For instance, mapping data from
in node
in node
, can represent
that the list of students in a university 1is the same to list of students in university 2.
Finally, data ecosystems can be enhanced with additional meta-data layers to enable
the description of business strategies and the access regulations. Figure 5 depicts the main
components of a network of data ecosystems empowered with these layers. As can be observed,
this enriched version of a network of data ecosystems comprises:
Meta-data describing business strategies will enable the definition of the stakeholders of
the network and their roles.
Objectives to be met and the dependencies among the tasks that need to be performed
to achieve these objectives.
3. Agreements for data exchange and criteria for trustworthiness.
4. Regulations for data access and for data privacy preservation.
5. Services composing services of the nodes of the network or exploiting their operators.
4.2.8 Research Challenges
This section presents an outlook on the main challenges to be addressed in the implementation
a network of data ecosystems.
4.2.9 Data and Metadata Curation
The availability and quality of data resources must be ensured, so that data value creation
can be stimulated. A promising solution is to use a well-conceived, efficient curation strategy
for data resources and their metadata. Such a curation technique is the continuous process of
82 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
managing, improving, and enhancing the data and their metadata. Furthermore, the curation
process aims to ensure that the data and metadata meet a defined set of quality requirements,
such as security rules, integrity constraints, or metadata availability expectations. Without
proper curation, data resources may deteriorate in terms of their quality and integrity
over time. One of the major challenges for achieving continuous curation of metadata is
to create a methodology to structure the curation process as well as to provide a set of
tools. Furthermore, data sets obtained through transformation processes of data analytics or
machine learning can become new data sets. Not only provenance but also the processing
methods should be associated with them.
4.2.10 Data Traceability and Data Consumption Monitoring
Tracking the utilization of data sets and applications using these data resources is still a big
challenge. Such information could be useful for both the identification of new data sets and
for data quality improvement. Monitoring data consumption, as well as providing effective
ways for the consumer to interact with the data publisher, should make it possible to collect
information about using and sharing data. In this sense, it is crucial to obtain consumers’
feedback in such a structured way that it allows identifying flaws in the published data, the
need to publish new data, and, for example, to enable classification of data.
4.2.11 Describing the Main Properties of Data Sets
All data sets in a data ecosystem should be described in terms of their main characteristics. In
particular, in the case of numerical data, a general framework that allows for the representation
of uncertainty and imprecision is required. In addition, keeping track of data set updates
and modifications, and allowing basic access to versioning information are also important
issues. For instance, a new version may be created when there is a change in the structure,
contents, or characteristics of a data set. As data sets can change over time, maintaining
different versions of the same data set and enable access to them becomes necessary.
4.2.12 Data Enrichment via a Joint Human/Machine Integration Effort
Data integration in a data ecosystem is challenged by the need to handle large volumes of
data, arriving at high velocity from a variety of sources, which demonstrate varying levels of
veracity. Data integration has been historically defined as a semi-automated task in which
correspondences are generated by matching algorithms and subsequently validated by a single
human expert. The reason for that is the inherent assumption that humans do it better which
is not necessarily the case. A current challenge involves the identification of respective roles
of humans and machines in achieving cognitive tasks in a way that maximizes the quality of
the integrated outcome. We observe that the traditional roles of humans and machines are
subject to change due to the availability of data and advances in machine learning.
4.2.13 Adaptive/Context-aware Data Quality Assessment and Redefinition of
Data Quality Dimensions
Data quality assessment requires the selection of both the dimensions to be evaluated and
the metrics to measure the selected dimensions. An accurate evaluation of the quality of the
data depends on diverse factors, e.g., the type of source or data, and the application that
aims to use the data. This implies an adaptive approach for data quality assessment able to
trigger the appropriate metric. Furthermore, data quality assessment should be performed
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 83
in two phases: the registration and the usage phase. During the phase in which data are
registered/ingested into a data ecosystem, a first evaluation is needed. This evaluation should
consider some basic metrics in order to guarantee that low quality data is not ingested in a
data ecosystem. On the other hand, in the usage phase, metrics that reflect the suitability of
a data set along the applications that aim to access the data set need to be defined. Moreover,
data quality can be measured at different levels. There are quality dimensions which refer
to the schema or overall structure of a data source. Additionally, data quality dimensions
regarding the content of the data sources can be defined. In both cases, metrics which
represent functions to determine the corresponding values for the data quality dimensions,
may cover atomic items, such as a single attribute, or multiple items, or complex objects. In
a single data ecosystem node, but especially in the case of sharing data between several nodes,
it is not clear, how data quality values should change, when the data is further processed.
The way to calculate new values for a single dimension might be dependent on the meaning
of the dimension and a given metric, as well as specifically on what happens to the data in
the current processing step.
4.2.14 Measuring Information Gain in a Data Ecosystem
Adding new data sets to a data ecosystem arguably results in some kind of information gain.
The gain extends naturally the increase in knowledge that occurs when new mappings are
discovered within a data ecosystem node – either interactively or semi-automatically, or
across diverse nodes in a data ecosystem. Questions that need to be addressed include:
i) How is the information gain defined and measured?
Can the gain be defined in terms of benchmark queries and the results they return given
an evolving state of a data ecosystem?
4.2.15 Exploring the Social Dimension of a Data Ecosystem
Users (humans, or data consuming services) are actors in a data ecosystem, playing an
important and active role in its evolution. Firstly, they can provide feedback on data sets
that they have used (either formally or informally), adding to the quality and provenance
properties, which other users should be able to take into account. Secondly, they may decide
to curate data sets that exhibit poor quality properties and return a better version of those
to the DE, generating more metadata in the process (provenance tracing with details of the
interventions). Finally, they may contribute to the mapping exercises that the DE makes
possible and indeed relies upon. Some of the questions that emerge when users are first class
citizens include:
Should data ecosystem users form a network, and if so, can such a network be used to
recommend / promote / deprecate data sets?
ii) Should users (or services) be credited for producing and making available new (better)
versions of a data set, i.e., through curation activities?
How can the social layer of the DE be promoted, maintained (is that a part of the
“upper layer”?), and exploited?
How would a recommender system be able to suggest data sets to users based on their
usage history of the data ecosystem and the implicit connections with other users?
84 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
4.2.16 Explainability in Data Ecosystems
Whenever the brain of a data ecosystem executes a service (e.g., query, data transformation,
quality evaluation, or mapping generation), it should be able to explain the result of the
operation carried out in executing the service using an appropriate language. There is a
connection between provenance and explanation. In some sense, provenance information can
be part of/exploited in an explanation. Included in the notion of explanation is the idea of
explaining the semantics of a data source–either primitive, or produced by a data operation.
4.2.17 Query Processing over Data Ecosystems
Effectiveness and efficiency of query processing over a federation of data sources is known to
be affected not only by the number of sources in the federation but also for the heterogeneity
of these sources. In case of a Date Ecosystem, data sources can be ingested in different
formats and accessible with management systems that provide different data management
capabilities, e.g., some systems may support complex queries including joins and aggregations
while others do not. Moreover, the data quality and structuredness may considerably vary.
For ensuring the correct processing of queries, interoperability issues should be solved at
different levels. Meta-data about the description of the meaning of the data stored in the
data ecosystem plays a crucial role for enabling the execution of queries over the data sources
that will provide the correct answers. Moreover, if several versions of the data sets are stored
in a data ecosystem, the process of data integration must take into account the temporal
dimension of data. For example, answering queries like tell me all the students of University
X who have become professors at X after no more than 10 years from graduation, might
involve looking at several versions of different data sets.
4.2.18 Data Ecosystems for enabling the specification of Meta-Metadata
The proposed model for Data Ecosystems enables the definition of a second layer of global
meta-data which describes both data and meta-data descriptions of all the local nodes in the
Ecosystem. Managing and keeping meta-metadata up-to-date is more challenging because
both data and metadata may change as a consequence of the dynamics of such systems.
Dynamicity may be at least twofold:
i) local dynamics of node data and metadata and
global dynamics of queries and mappings involving several nodes and the Data Ecosystem
metadata themselves.
The research questions to answer under these conditions are:
a) How to define safety criteria?
b) What kind of transformations should be allowed?
c) Should we prevent non-conservative structural updates?
d) Which level of data integrity and consistency will be ensured?
4.2.19 Specification language for procedure synthesis
Beside ontologies and meta-data, a language for specifying services and their functional
properties is needed. This is not just for the sake of verification, but mainly because they
are required for procedure synthesis. For automatic search through large libraries and for
synthesis to be feasible such a language should balance expressive power with effectiveness
and efficiency. Attempts to use type theory for these purpose include work by Bessai et al
[10] and Hengelin and Rehof [58].
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 85
4.3 The Business of Data Ecosystems
Elda Paja (IT University of Copenhagen), Matthias Jarke (RWTH Aachen &Fraunhofer
FIT), Boris Otto (Fraunhofer ISST &TU Dortmund) and Frank Piller (RWTH Aachen)
License Creative Commons BY 3.0 Unported license
©Elda Paja, Matthias Jarke, Boris Otto, Frank Piller
4.3.1 Background and Motivation
IExample 1
(Data Ecosystem)
Consider the example of PRINT INC., an industrial print
shop reacting to increasing demand by adding more flexible production capacity. The firm
provides machine vendor DRUCKMASCHINEN AG a list of requirements, derived from
its most sophisticated print jobs. DRUCKMASCHINEN, however, lacks insights about the
specific situation and suggests a standard specification. After ramp-up, a sample analysis
reveals an OEE far below the anticipated values, as configurations, parameters and procedures
of the existing facilities do not recognize the capabilities of the new equipment. This scenario
is still a very common situation in many industries. With a data ecosystem in place, PRINT
INC. could first provide DRUCKMASCHINEN better past operating data to better determine
requirements. A specialized analytics service provider would assist in this analysis, also
drawing on technology insights and usage data from other print shops operating in a related
setup. After implementation, a higher level of OEE could be reached by learning from best
practices from similar operations of other manufacturers. By getting access to rich usage
data – covering not just machine data, but also data on the production context, material
characteristics and behavior of PRINT INC – OEE can be continuously improved. The data
ecosystem enables a continuous feedback of this data into the development cycle of DRUCK-
MASCHINEN, revealing requirements for the next generation of production hardware and
software. In times of low capacity utilization, PRINT INC. provides DRUCKMASCHINEN
access to its new printing machine, which becomes part of DRUCKMASCHINEN’s pro-
duction network, connecting printing capacities from hundreds of printers virtually. With
its PRINT HUB platform, DRUCKMASCHINEN can place print jobs from clients like
publishing houses (without own production capacity) on existing machinery, generating a
new business opportunity for both itself and its customer. At the same time, by moving
from the role of a manufacturer and service provider of printing machine to that of an
operator, DRUCKMASCHINEN also gains critical knowledge in the operating principles and
continuous improvement opportunities of its machinery.
To realize this typical vision for Industry 4.0, much more than a technical infrastructure
is required. Beyond the availability of data and capacity, this scenario only can be realized
if it has been determined how the value gained by accessing these inputs is captured and
shared among the actors involved:
Would profiting from such external data also require to provide similar feedback data?
Could PRINT INC. control who and what is printing on its capacity shared via PRINT
How could print shops differentiate themselves by process know-how when a data ecosystem
enables a kind of instant benchmarking, as all operational data is shared over the network?
Why is DRUCKMASCHINEN, in the end, not operating all its machines directly, integ-
rating vertically into the domain of its former clients?
Who owns the governance of this network? Is it DRUCKMASCHINEN as the focal
keystone player, or should rather an alliance of print shops, users (like publishing houses)
become the owner of the platform [113]?
86 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
This generic example provides a glimpse of the challenges from an economic and com-
petitive dynamics perspective that come with the vision of a data ecosystem, which is, by
definition, not restricted to a focal company or value creation within a closed network of
established partners. It resembles the vision of an open network of sensors, assets, products,
and actors that continuously generate data. This data is utilized to enhance operational
efficiency, but also to provide new opportunities for strategic differentiation. A core element
hence is a business-model perspective on
(re-)usage of data, insights, and applications
by other parties
than those generating the data at the first place. Generally, learning and
analytics can take place faster and more efficiently if manufacturers not only utilize their
own data, but also could access data from similar contexts in other industries.
Actual research on industrial data-based business models has focused predominantly on
the perspective of value creation, i.e. how to use shared data to create new service offerings like
predictive maintenance, energy optimization, quality improvements, etc. [
]. But the open
question still is an understanding how
to incentivize the sharing of deep production
know-how and data
in order
to balance value creation (using the data) with value
capture (sharing the rents)
]. The
rise of platforms
where these data are being
exchanged and enhanced by dedicated “apps”, often offered by specialized third-party entities,
one of the largest economic developments of the last decade
]. As platform
become more open, more actors will be attracted into a data ecosystem, and
the platform orchestrator will be able to access a larger set of potentially complementary
innovative and operative capabilities. Most of the external contributors will innovate in
ways complementary to the platform. Some, however, may start developing
in ways that become competitive to the platform. Such an emergence of competition will
depend on the
governance of the ecosystem
. Collaborative governance mechanisms will
increase complementors’ incentives to innovate in platform-enhancing ways. This demands
especially a dedicated setup to share the value created on the original platform, but is also
dependent from the technical interface design. In turn, emergence of competition from former
complementors is likely to create a reaction by the platform leader to start competing back
with these new rivals, either by enveloping them, or by closing its technological interface,
in effect
moving away from being an open data ecosystem
towards becoming a more
closely managed network or internal platform. Based on a discussion of different governance
frameworks for data sharing and access, this workgroup of the
Dagstuhl Seminar #19391
discussed various scenarios, but especially questions for further research. Considering actual
developments in platform-based business models that recently emerged also in manufacturing
industries, our workgroup discussion tried to understand the design of the business models,
but especially how to incentivize (reward) the sharing of deep production know-how and
data in order to balance value creation (using the data) with value capture. This led to the
following questions suggested for future research:
Modeling the tension between openness in value creation and control of value capture,
while recognizing the need for establishing and increasing trust in data sharing.
Managing property rights (access, transfer, enforcement) at data, applications, and
connected assets as a result of varying degrees of platform openness
Definition of governance modes and design factors to generate adequate business models
for a data ecosystem that allow to maximize value appropriation for all involved actors
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 87
Figure 6 Digital Business Ecosystem Example.
4.3.2 From Data Ecosystems to Business Ecosystems
Figure 6 illustrates the interactions within a digital business ecosystem in terms of data,
cash and payment flow. Within the ecosystem perspective, a holistic view on the business
relations is provided.
The digital ecosystem is enabled via multiple technical infrastructures. In this example
two major infrastructure providers (data marketplace operator and software vendors provide
a network to connect various entities such as the physical machines, customers, service staff,
and third party services e.g. for data analyses. The Original Equipment Manufacturer
(OEM) collects data through specially installed edge devices on the customer’s sites. The
operational machine data is transferred to the platform-based cloud infrastructure provided
by the OEM. The data is stored, processed, and analyzed on the OEM’s platform. Part of
the data analysis is performed by the OEM itself within separate organizational units, so
that the role of the data analytics provider is also taken internally. Therefore, additional
historical and product development data is provided from internal resources to improve the
analysis. Based on the analyses of the OEM, various services can be provided to the plant
operators via the platform (e.g., dashboards on machine usage, specific operational reports,
process optimization recommendations, maintenance planning). New external data sources
(e.g., weather data) from meteorological stations can be considered as new parameters for
analysis. This data is provided within a marketplace, which the OEM uses to enrich the
existing data basis. The data marketplace acts as a gateway to other data ecosystems and
promotes cross-industry data exchange. The ecosystem perspective provides an overview of
the dynamics within a practical data ecosystem including the flow of data, payment and
88 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
4.3.3 State of the Art
One of the most important questions of a firm is how to efficiently create value: Should it
produce its own output or should it orchestrate the output of others? In the case of software,
the choice increasingly favors orchestration over production. Apple, Google, and Microsoft
became the three most valuable companies in the world in 2015 by relying on a
business model
to provide software applications for its users, developed by independent
app programmers. An open platform business model offers distinct economic advantages
because it allows a firm to harness external inputs and innovation as a complement to
internal innovation [
]. These platform markets are, however, neither a new phenomenon
nor restricted to software or information goods. Open platforms have emerged in aerospace
(Lockheed Martin), T-shirts (Threadless), 3D printing (MakerBot), and shoes (Adidas).
Since the early 2000s, the industrial organization literature has begun to develop theory
on platforms (also referred to as “two-sided markets”, “multi-sided markets”, or “multi-sided
platforms” [
]). Economists view platforms as special kinds of markets that play the
role of facilitators of exchange between different types of users that could not otherwise
transact with each other. Essential to most economic definitions of
multi-sided platforms
(MSPs) are the existence of “
network effects
” that arise between the “two sides” of the
market [
]. As the value of the platform stems principally from the access of one side to the
other side of the platform, the question of platform adoption becomes how to bring multiple
sides on board. Such platforms typically reside upon a
layered digital infrastructure
where lower-level layers (e.g., physical components, transmission layer) enable and support
functionalities at higher, user-facing layers (e.g., operating systems layer, application layer)
]. To create value, ecosystems hence depend on complementary inputs made by
loosely interconnected, yet independent stakeholders from varying levels of (technological)
distance from the end consumer [3, 90].
facilitate the establishment of business ecosystems, which are formed by the users
interacting on the platform. Researchers have examined a number of case studies on MSPs in
which a keystone firm owns and governs the platform [
]; so these are “keystone-driven
MSPs”. However, the platform landscape is becoming more and more diverse, with other,
more complex governance and ownership structures being observable in different domains. De
Reuver et al. [
] have found that in many cases there is no single platform provider, but that
the platform is jointly designed and “shaped” by multiple actors. In this context, Tiwana et
al. [
] point to the importance of distinguishing platforms owned by a single firm from
platforms characterized by some form of “shared ownership”. Shared ownership materializes
in multiple organizational forms, among them the alliance. Gawer [
], for example, identified
some MSPs from the supply chain management domain that are “shared among firms that
are part of a formal alliance”. Such “alliance-driven MSPs”, which are characterized by
shared ownership and governance (e.g. a joint-venture company or industry association), as
well as decentral platform governance models have been neglected by academic research so
far [30].
Table 1 compares keystone-driven with alliance-driven MSP design.
In the case of an
industrial data ecosystem
, platform participants include the
chestrator of the platform
(today either IT infrastructure providers like SAP or Software
AG, or providers of automation or manufacturing equipment like Siemens, GE, Trumpf),
operators of
production assets
(users in form of “factories”), which provide data, providers
4This paragraph is taken from Otto and Jarke (2019) [89]
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 89
Table 1 Juxtaposition of Keystone-Driven and Alliance-Driven MSP Design.
Theoretical Concept Keystone-Driven Alliance-Driven
Platform architecture
Architecture determined by
goals of keystone firm
Architecture determined by shared in-
terest of multiple owners (leading to
decentral data storage, for example)
Platform boundary
Mainly technical boundary
resources (APIs, SDKs etc.),
supported by “social” bound-
ary resources (e.g. training
for developers) [14]
Data (IDS specific) as a boundary re-
source of “dual” nature, i.e. requiring
both technical processing and functional
use; many social boundary resources,
such as working groups, task forces etc
Platform design
Core developed by platform
owner, then extended by
Consensus oriented design process with
focus on “common denominator”
Platform ecosystem
1) Innovation, 2) Adoption,
3) Scaling [57] 1) Adoption, 2) Innovation, 3) Scaling
Start with limited number
of sides and limited options
for interaction between them,
then increase number of sides
and options for interaction
Start with complex ecosystem (i.e.
multi-stakeholder setting), then reduce
to core ecosystem and extend it later
on depending on roll-out requirements
Mainly pricing instruments,
accompanied by non-pricing
Dominated by non-pricing instruments;
integration of pricing instruments sched-
uled for scaling phase; data governance
of applications analyzing this data and providing prescriptions and predictions (
app pro-
), and other asset operators that utilize insights from aggregated data to optimize
their own production. In addition, also the goods being produced can become part of the
platform in form of
connected (“smart”) products
, providing a feedback loop of usage
data, but also becoming the center of another platform around digital services complementing
these products. The latter also refer to
end-users (customers)
as a final participant of
the ecosystem.
An applied stream of research looked into the demand for business model innovation
(BMI) in leu of Industrie 4.0 [
]. Based on rather qualitative research approach, a
demand for professionalizing the BMI process in established companies was identified [
Similarly, specific aspects of BMI for Industrie 4.0, i.e. specific patterns or components of I40
BMs, have been identified. By analyzing Industrie 4.0 characteristics for a firm’s business
model and the approaches of pioneering companies to instigate BMI, this research developed
a methodology to both generate, model, systematically design, and evaluate BM alternatives
triggered by Industrie 4.0 [93].
Another recent stream of literature complements the original economic analysis of plat-
forms with a more managerial and design-orientated perspective. It looks into the distinct
governance and orchestration challenges
presented by established innovation ecosys-
tems (e.g., [
]). This literature has looked, for example, into prices, incentives, contracts,
and network effects. Less work has addressed how prospective ecosystem stakeholders
mit resources
towards a de novo ecosystem creation effort and how they evolve a shared
structure of interactions [
]. Yet, this is exactly the situation of many data ecosystems,
where creation is not a simple endeavor of an app. Rather, the ecosystem value proposition
depends on the concurrent availability of complementary inputs from varied, independent
90 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
In such a situation, a core decision of a platform operator is about how much
to open the
and when to absorb inputs (developments, apps, data) from the connected parties.
This decision drives adoption and harness developers as an extension of the orchestrator’s own
production function [
]. [
] extend the analysis to contexts in which the establishment
of a new business ecosystem cannot be reliably planned, as no clear value proposition to
orchestrate the ecosystem exists ex-ante. In a piece of complementary research, [
] investigate
the perspective of platform complementors (app programmers) and their decision to join
a platform based on its openness. The authors develop and validate a platform
measurement instrument that captures perceived platform openness.
Since 2014, our Fraunhofer Industrial Data Space initiative [
] has focused on
requirements and rather technological challenges of inter-organizational data exchange.
This requires novel conceptual information modeling at multiple levels, and still significant
research for the specialization to the case of production engineering. The ideas pursued
at the boundary of CRD_A and this workstream build on significant prior research on
the conceptual modeling, structure and evolution analysis, and data-based Social Network
Analytics concerning strategic dependencies and trust among players [46, 47, 84].
4.3.4 Conceptual Background5
The proliferation of digital technologies and AI accelerates a shift in business models which is
characterized by an increasing importance of data as a strategic resource. While traditional
business models rest on tangible assets, data is the raw material not only for information
and knowledge, but for innovative services and customer experiences. Besides the shift
from tangible to ”smart” products and from controlling the physical to orchestrating the
data value chain, there is one additional fundamental change in the digitalized economy.
Innovation increasingly takes place in ecosystems in which various members such as businesses,
research organizations, intermediaries such as electronic marketplaces, governmental agencies,
customers, and competitors band together to jointly achieve innovative value offerings.
Ecosystems are characterized by the fact that no one member is capable of creating
innovation on its own, but the ecosystem as a whole needs to team up. In other words:
Every individual member has to contribute in order to benefit. Ecosystems function in an
equilibrium state of mutual benefits for all members.
In a data ecosystem, data is the strategic resource for the success of the whole system as
it is understood as a stand-alone asset that will be exchanged and monetarized within the
ecosystem. That offers the participating actors new growth opportunities through networking
with other participants and acts as a driver for innovative services and customer experience.
Data sharing opens up new opportunities for progress and the formation of cooperations
with other companies or actors from which every participant in the data ecosystem benefits.
The various activities of the different members in a data ecosystem lead to a complete
coverage of the data value chain. This includes the stages of data generation, curation,
exchange, storage and analysis as well as the use of the resulting knowledge for comprehensive
business decisions. Through the sustainable data exchange the participating actors are
able to develop further and to operate value co-creation which leads to new digital value
5Cf. Otto et al. (forthcoming). [88]
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 91
4.3.5 Data as a Boundary Resource6
As far as the boundary resources are concerned, Henfridsson and Bygstad [
] argue that
this concept is helpful for studying patterns of interaction between the various groups and
agents on a digital platform. Boundary resources are resources through which different
agents create relationships and interact with each other in order to co-create value [
]. Dal
Bianco et al. [
] distinguish between technical and social platform boundary resources.
Typical boundary resources are Application Programming Interfaces (APIs) and Software
Development Kits (SDKs). Examples for social boundary resources are intellectual property
rights and documentation of software services. Furthermore, boundary resources are not
stable, but evolve over time. Eaton et al. [
] coin the notion of “distributed tuning” to
describe the process of continuous shaping and reshaping of boundary resources between the
different platform actors and users. More recent research has suggested to increasingly look
at such boundary resources of digital platforms as a promising subject of analysis [30].
How organizations can exchange and share data has long since been an important research
topic. The need for companies to exchange and share data has been a major motivation for the
development of platforms mediating between suppliers and buyers of goods. Early two-sided
data exchange solutions were facilitated by technological standards, such as EDIFACT or
ANSI X.12. Gawer [
] within her integrative platform framework identified traditional
buyer-supplier relationships for which data is a technical enabler.
Around the turn of the millennium, electronic marketplaces emerged as intermediaries to
reduce the complexity of the increasing need of n:m data exchange [
], in which data from
multiple sources (n) can be bundled and utilized in contextualized presentations to multiple
users (m). This intermediary function comprised – among other things – the mapping of the
different message schemas of the various standardization initiatives that evolved. Motivated
by the success of peer-to-peer-networks in the consumer realm, some researchers explored
technologies, and even business models, for peer-to-peer based networks for data exchange
in the industrial domain. Technological aspects of peer-to-peer data ecosystems, such as
context exchange among different world views of organizations [
] or automation of data
mappings in heterogeneous settings [
], have been investigated since the late 1990s. In 2005,
Franklin et al. [
] noted the growing richness of digital media and proposed that users
should be enabled to create their own “data space”, where a free collection of data and media
objects could be managed under a user specific network of semantic metadata. In 2010, the
notion of the “data lake” was coined [
] and quickly received attention in the practitioners’
community. Furthermore, some researchers investigated the role of data within platform
based ecosystems [
]. More recent research has dealt with the upcoming phenomenon
of data platforms, mainly encouraged by the discourse around big data [
]. These
studies focus mainly on platform architecture technology and data flows.
4.3.6 Research needs
While the emergence of data ecosystems offers new opportunities for the different ecosystem
participants, many social, environmental and business challenges have to be addressed in
order to pave the way for these opportunities to materialize. Among the most significant
challenges are:
6Section taken from Otto and Jarke 2019 [89]
92 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
Trust: New methods are needed to increase trust in data sharing so that more data would
be available for new applications. What is needed is a framework that includes building
blocks for data sharing, data management, data protection techniques, privacy-preserving
data processing and distributed accountability and traceability. In addition to providing
technology for platform developers, the framework should provide incentive and threat
modelling tools for data sharing business developers and strategists, who consider opening
data for new cooperation and business.
Data Sovereignty: The framework should also support compatibility with the latest and
emerging legislation, like the EU’s General Data Protection Regulation (GDPR) and free
flow of non-personal data, as well as ethical principles, like IEEE Ethically Aligned Design.
This will increase trust in industrial and personal data platforms, which will enable larger
data markets combining currently isolated data silos and increase the number of data
providers and users in the markets. The result should aim to be platform-agnostic to be
applied in multiple domains with platforms based on different technologies.
Interoperability: The main objective should be to support a trusted data ecosystem
providing easy-to-use privacy mechanisms and solutions that guarantee citizens and
business entities can fully manage data sharing and privacy. The challenge is thus to
provide a corresponding overall technical architecture which needs to take into account
the key reference platforms and technologies to support data sharing, to improve existing
solutions and architectures, to define the overall reference architecture, and to design
platform-agnostic trusted data sharing building blocks and interoperability.
Data Governance: Data Ecosystems highly depend on access to data and interactions
of actors providing or using data or similar resources such as application programming
interfaces (APIs). The role of data governance in these complex networks between
organizations is an under-researched field. There is a lack of concepts and mechanisms
to mandate responsibilities among participants of a data ecosystem. It is essential to
study inter-organizational mechanisms that allow participative interactions, incentives to
influence the dynamics and evolution of the ecosystem.
Compliance with Antitrust Legislation: To avoid the risk of data monopolies, the following
needs to be ensured:
Improving the mobility of non-personal data across borders in the single market, which
is limited today in many member states by localization restrictions or legal uncertainty
in the market;
Ensuring that the powers of competent authorities to request and receive access to data
for regulatory control purposes, such as for inspection and audit, remain unaffected;
Making it easier for professional users of data storage or other processing services to
switch service providers and to port data, while not creating an excessive burden on
service providers or distorting the market.
Data Economics: Data business, i.e. viewing data as an economic asset, will bring
additional motivation for data providers and owners to open up their data for various
applications. Personal data is becoming a new economic asset class, a valuable resource
for the 21st century that will touch all aspects of society. The rapid development of the
Personal Data Service (PDS) market will provide big changes in the way individuals,
business and organizations deal with each other, as individuals assert more control over
their data or service providers process personal data.
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 93
4.3.7 Data Infrastructures
A data infrastructure is a distributed technical infrastructure consisting of
components and services which support data access, storage, exchange, sharing and use
according to defined rules.7
International Data Spaces (IDS)8
initiative aims at data sovereignty for busi-
nesses and citizens in Europe and beyond. The IDS Association (IDSA) provides a reference
architecture that enables an ecosystem for the sovereign exchange of data with clearly defined
usage rights. The reference architecture defines a technical infrastructure and includes
contractual regulations: at the semantic level, data linking, or analysis can technically
be prevented or made possible. In this way, the classic structure of cloud services is also
embedded in an interoperable digital economy with full data sovereignty on the digital
infrastructures of third parties. The IDS standard solves a market obstacle: In order for
data to unfold its value creation potential, it must be described and tradable according to
a global and interoperable standard. This has never existed before. But DIN SPEC 27070
(to be published in Nov 19) is the first global and interoperable standard. 100 member
institutions from the EU as well as from Brazil, Canada, China, India, Japan and the USA
are involved; they come from all branches of industry and have developed information and
governance models in the IDSA as the basis for the IDS architecture and its data sovereignty
standard. More than 50 concrete application scenarios and first products are now available
from companies of all sizes and sectors – together they are working on operational concepts for
a sovereign data exchange infrastructure. The certification scheme “IDS_ready” also enables
companies outside the association to participate in secure, IDS-based value-added processes
via certified participants and components. Reference implementations and sample codes are
available for developers and can be tested in testbeds. IDSA is in continuous coordination
with global initiatives (Industrial Internet Consortium, OPC Foundation, Robot Revolution
Initiative, BDVA) and participates in EU research projects to anchor IDS architecture and
data sovereignty standards within European digitization strategies. In 8 countries, there are
contractually bound IDSA hubs that bring standardization and adaptation of the technology
to the respective country.
4.3.8 Questions for further research
From this brief review of literature in the field, but especially our conclusions during the
seminar, we draw three conclusions:
The basic mechanisms of platform markets are well understood, especially the basic effect
of network effects and complementing value creation in multi-sided markets.
Platform openness has derived as a key variable in the systematic design of a platform
ecosystem. Literature suggests some factors that constitute openness that can be utilized
for the design of a data ecosystem.
All existing analysis, however, has been conducted in the field of either consumer elec-
tronics or information industries (gaming, search engines, social media sites). Dedicated
research in the context of industrial data applications is missing.
For the source of this definition, please see
Digitale-Welt/das-projekt-gaia-x.pdf?__blob=publicationFile&v=18 (in German, English version to
follow very soon).
For the source of this section, please see
94 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
Future research in the field is required in various aspects, which can be structured in four
layers of analysis, as suggested by Gawer [49]:
Interfaces: On a technical level, the openness of APIs and other technical interfaces is
not just a question of programming and quality control, but first an important design
factor for the ability of connected asset to perform predictive and prescriptive functionality,
i.e. to enhance its capability in this regard by getting access to data also from other actors.
From the perspective of a platform, the openness of an API is a signal of willingness to
share data and knowledge, hence potentially attracting third parties. At the same time,
open interfaces are not just a technical risk, but also reduce the ability of the originator
of the data to capture unique value from this data and hence differentiate it from other
market players. Research needs to investigate the choices made by companies on different
levels of a platform to understand the decisions and their consequences with regard to
interface design. This also includes the deployment of Software Development Toolkits
(SDK) by a platform operator, which determine the ability of third party application
: This layer deals with capabilities that organization need to acquire to
position themselves in an data ecosystem and corresponding platforms. These capabilities
include business model innovation, mastering organizational change, or capabilities of
orchestrating a manufacturing ecosystem. This research builds on a rich literature of
capability building and organizational sense making and will study how dedicated capab-
ilities link to firm performance. Of particular interest are question of counterbalancing
(the lack of) capabilities of one actor by capabilities of another actor on a platform. This
leads to a re-interpretation of the central economic question of the boundaries of a firm.
Organizational design
refers to the design of a platform ecosystem and the design
factors (“business model patterns”) that allow for value creation and capture in these
industrial data platform. A particular focus of this research will be on the level of value
capture, i.e., on mechanisms that allow the different actors of a platform for profit from
their participation and contributions of the platform. Building on the last work package,
this also asks the question whether firms shall joining an existing ecosystem (and of,
which one and under which conditions) or try to orchestrate an own?
Governance modes
: This final layer integrates the previous work on platform-based
value creation and value capture from an external, but also an internal perspective. A
central construct in this work is the degree of openness vs. desire for control of each
actor in the ecosystem. Future research needs to identify possible governance modes (and
patterns of platform governance) and match those to the performance of observed use
cases. This should help us to understand the choices made by managers in setting up
these governance modes and will (for example, game-theoretical) model the theoretical
consequences of the choices under given contingencies.
4.3.9 Outlook: How to compete when all become the same (have access to the
same data)
The use of networked and intelligent production systems and dynamic value creation networks
in the context of a data ecosystem accelerates a faster exchange of “best practices” across
corporate boundaries. For instance, this can be attempts to the process optimization in
production networks, but also access to the same complementary service of a platform provider.
Thus, operational efficiency can be increased for an entire industry. However, competitive
advantages do not result from operational efficiency, but from strategic uniqueness! As a
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 95
result, existing business models will be challenged. Companies that are highly focused on
operational efficiency are facing increasing competitive pressure. Efficiency gains cannot
be narrowed down to just one company. Instead of being entirely focused on operational
efficiency, business models for data ecosystem must therefore initiate new differentiation
opportunities for companies. The options for a more efficient and scalable custom performance
have already been addressed. Openness and transparency can also become a differentiating
feature: Not only during the innovation process, companies need to act more open instead of
cutting themselves off. This also includes an intelligent and transparent handling of data.
This intelligent handling of data enables the development of differentiating potentials using
customer-specific knowledge. For example, this can be used to generate additional services
or new products that satisfy the customer’s benefit even better. Transparency and fairness
can thereby become a competitive advantage. For companies, the challenge is to reach and
maintain the competitive position as leading innovator in a specific industry. This can only
happen with the best possible mix of open innovation processes and internal innovation.
Companies need to find an implicit or explicit trade-off with regard to the openness of
innovation processes and the type of shared knowledge. This includes finding strategies of
how to implement innovations faster to the companies without internal barriers causing
significant delays.
5 Use Cases
5.1 Use Cases from the Medical Domain
Sandra Geisler (Fraunhofer FIT – Sankt Augustin, DE), Maria-Esther Vidal(TIB – Hannover,
DE), Elda Paja (IT Univerity of Copenhagen), Maurizio Lenzerini(Sapienza University of
Rome, IT), Paolo Missier (Newcastle University, GB)
License Creative Commons BY 3.0 Unported license
©Sandra Geisler, Maria-Esther Vidal, Elda Paja, Maurizio Lenzerini, Paolo Missier
In the health care domain a variety of use cases exist, where data ecosystems are beneficial
and open up new opportunities via data exchange. In the following, we describe two use
cases which build complex data ecosystems in detail and analyze the challenges which are
posed by them and other use cases in the health domain to data ecosystems.
5.1.1 Use Case: Multi-Site Clinical Tríal
In this example application for a data ecosystem, in the course of the research project
SALUS (Selbsttonometrie und Datentransfer bei Glaukompatienten zur Verbesserung der
funded by the Federal Ministry of Health, a multi-site clinical trial
with glaucoma patients is conducted over one year in Germany where about 2000 patients
will be included in the study starting in 2020. Glaucoma is a chronic disease of the eye
with various causes possibly leading to irreversible damages of the eye nerves. For glaucoma
patients it is crucial to keep the intra-ocular pressure in certain bounds to not exacerbate
the condition which may lead to blindness in the worst case. The usual diagnostics require
the regular creation of a status quo of the patient’s condition, including an intra-ocular
96 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
pressure profile over two successive days. For this procedure the patient has to be admitted
to hospital. In the SALUS trial the advantages and disadvantages of using a mobile device
at home to create a pressure profile over one week, called self-tonometry, is compared to
the method applied in hospital. Additionally, the patient is equipped with a 24-hour blood
pressure device, and at examinations questionnaires will be completed by the patient.
5.1.2 Study Process
Patients are included in the study by a local ophthalmologist who will explain the study
to the patient and assign her to either the self-tonometry group or a control group ran-
domly. Successively, the patient will be sent to a local hospital which will do the screening
examinations and a study nurse will train the patient in using the self-tonometer. After the
examination, one week of self-tonometry or a two day hospital stay, respectively, follow. The
current therapy is adapted by the local ophthalmologist based on the results, if necessary. At
regular intervals follow-up examinations are executed by the local ophthalmologist. After 12
months a final examination is done in the local hospital. After the trial, the collected study
data will be evaluated by a research institute combining it with additional data provided by
the health insurance companies of the patients.
5.1.3 Data Processing
During and after the trial all collected examination data, device data, and images of each
patient have to be available especially to the local ophthalmologist, but also to the local
hospital, treating this particular patient. They need the data to analyze the current condition
and adapt the therapy where necessary. Hence, a glaucoma health record is maintained
for each patient at a central server. Data from the self-tonometry and the 24-hour blood
pressure monitoring device is transferred via a mobile device to the corresponding device
manufacturer’s portal. From there it is transferred anonymized to the trial server. The data
and images from the examinations by the ophthalmologist and hospitals are also transferred
from their systems and devices via a web application in an anonymized way to the central
server. After completion of the trial, the data from the trial server has to be integrated with
data of the health insurance companies and transferred to enable a comprehensive analysis
by a research institute.
An overview of the use case, the involved nodes and stakeholders, and the corresponding
dependencies aligned with the DE model from Section 4.2 are presented in Figure 7.
5.1.4 Challenges
We identified multiple challenges for this use case, which may also be generalized to other
data ecosystems in the health domain. By the time writing, the project is still in an early
stage, but where applicable, we sketch strategies for tackling the challenges.
Data views and access control.
For each stakeholder, different views on the data have to
be provided, as every role in this complex scenario has different access rights to the data.
Hence, a very fine-granular definition of access control rights has to be provided to respond
to this requirement. The data from the local hospitals, local ophthalmologists, and the
device manufacturer services has to be anonymized and integrated to provide the data for
the different views mentioned at different points in time during and after the trial.
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 97
Patients Ophtalmologists Hospitals Health Insurance
Companies Study Nurses Consortium
Manufacturers Strategy / Business
Properties Descriptions Standards Domain
Data Analytics Data Visualization Search
Regulations: GDPR, Trial Regulations, Data Protedction Laws
Data Operators
Descriptions StandardsF
Mappings: Annotations with Domain Ontologies
Strategy/Business: Hospital Management
Regulations: GDPR, hospital
regulations, national & federal protection
laws, trial regulations
Read Update Write
Node Local Clinic 1
Images Questionnaire
Data Sources
Node Local Opthalmologist
Node Device Manufacturer
Node Health Insurance Company
Data Operators
Descriptions StandardsF
Mappings: Annotations with Domain Ontologies
Strategy/Business: Trial Management
Regulations: GDPR, national & federal
data protection laws, trial regulations
Read Update Write
Node Glaucoma Trial Server
Process Data Master Data
Data Sources
Mappings: between data and
metadata for data Integration
Mappings: between data and
metadata for data integration
Mappings: between data and
metadata for data integration
Figure 7 Overview of the Multi-Site Clinical Trial.
Data Integration.
Thus, we need an integrated patient-centric global schema for a glaucoma
health record where all data from each patient identified by a global identifier is combined.
It has to be considered, if this integration should be virtual or materialized, and, how the
mappings between the global schema and the sources can be created. In this use case most
likely materialized integration will be used as the main data sources will not change frequently.
Further integration of external data, such as medication information or usage of standards
for enrichment and documentation, is a challenge given that suitable sources have to be
found, their quality has to be checked, and the corresponding mappings have to be created.
In this scenario various heterogeneous data sets and standards are involved which makes the
integration a challenge. A data lake as basic data management architecture for the health
record could be a solution to integrate various heterogeneous data sources and also allow
for complex meta-data management, data quality management, access control, and search
on the data. For each party/view a separate data mart could be created as a subset of the
data, strictly controlling the access to the data. But such an architecture would imply a high
implementation overhead as mature data lake systems are still not the usual case. A more
simpler solution to be flexible according to the data storage could be also a single NoSQL
database, where meta-data management and data quality management have to be realized
Consent management vis-a-vis the GDPR.
A further challenge is the implementation of
consent management and usage control in compliance with the GDPR. Corresponding means
for the deletion or editing of data at one or multiple available sites have to be implemented if
the patient revokes the consent or asks for corrections. Furthermore, the GDPR also demands
for transparency of operations on personal data, which requires a form of provenance tracking
and auditing. Data protection and secure communication also have to be ensured, when
data is exchanged between the nodes. Finally, quality monitoring during the trial can be
implemented to get high value data.
98 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
5.1.5 Use Case: Precision Medicine and Health Policy Making.
In this case study, the focus is on the integration of structured and unstructured data
ingested from clinical records, medical images, scientific publications, or genomic analysis.
The application of a network of data ecosystems is illustrated in the context of the EU H2020
funded project iASiS
which aims at exploiting Big data for paving the way for accurate
diagnostics and personalized treatments. iASiS
is a 36-month H2020-RIA project that has
run from April 2017 to March 2020, with the vision of turning clinical and pharmacogenomics
big data into actionable knowledge for personalized medicine and decision making. iASiS
aims at integrating heterogeneous Big data sources into the iASiS knowledge graph. As
input of the problem, we have a set of myriad sources of knowledge about the condition of a
lung cancer patient. Electronic health records (EHRs) preserve the knowledge about the
conditions of a patient that need to be considered in order to have effective diagnoses and
treatment prescriptions. Albeit informative, EHRs usually preserve patient information in an
unstructured way, e.g., textual notes, images, or genome sequencing. Furthermore, EHRs may
include incomplete and ambiguous statements about the whole medical history of a patient.
As a consequence, knowledge extraction techniques are required to mine and curate relevant
information for an integral analysis of a patient, e.g., age, gender, life habits, genotypes,
diagnostics, treatments, and family medical history. In addition to evaluating information in
EHRs, physicians rely upon their experience or available sources of knowledge to identify
potential adverse outcomes, e.g., drug interactions, side-effects or drug resistance. Diverse
repositories and databases make available relevant knowledge for the complete description
of a patient’s condition and the potential outcome. Nevertheless, sources are autonomous
and utilize diverse formats that range from unstructured scientific publications in PubMed
to repositories of structured data about cancer-related mutations. Various services need to
be implemented in order to transform relevant data that come in different formats into a
common format and for data anonymization. Additionally, since the same concept can be
identified with different identifiers, the detection and representation of mappings between
data sets is necessary. Each data provider establishes a regulation about the type of operators
and services that can be executed over each data set; they can also indicate which of the users
of the data ecosystem can execute the operators and the services. Finally, GDPR regulations
need to be respected ensuring that personal data is used according to the consents provided
by patients. Figure 8 illustrates an overview of a solution precision medicine use case using a
network of data ecosystems.
5.1.6 Challenges
Clinical data is usually stored in diverse formats, e.g., in notes in Clinical Records, gene
sequencing panels, or medical images. Additionally, these data sources may suffer from
potential biases, ambiguities, and noise. To overcome these data issues, distinct knowledge
extraction methods need to be included as part of the services of the network. Typical
extractions methods include:
i) Natural language processing (NLP);
ii) Visual analysis and image processing; and Genomic Analysis.
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 99
Clinical Data Sources
Data Operators
Descriptions Properties Domain Ontologies
Mappings: Annotations with
Domain Ontologies
Data Operators
Descriptions Properties Domain Ontologies
Mappings: Annotations with
Domain Ontologies
Mappings: Between entities annotated with the same terms
Descriptions Properties Domain Ontologies
Strategy/Business Strategy/Business
Regulations: GDPR
Node Clinical Data Node Scientific Publications
Regulations: Partners 1 and 2 access data
sources and the services and GDPR
Regulations: Open Source
EHR Genomic Images
Anonymity NLP Genomic
Patients Hospitals Physiancs Research
Partners 1, 2, 3, 4
Read Update Write Read Update Write
Search Engine
Analytics Descriptive
Figure 8
Overview of a Biomedical Data Ecosystem. A Network of Data Ecosystems for the
Precision Medicine Use Case; it comprises two nodes of clinical data and scientific publications.
Additionally, data quality assessment techniques are required in order to identify data quality
issues, as well as to define strategies for data curation. Data integration taking into account
the meaning of the biomedical concepts is also challenging as the consequence of the variety of
formats and representations. Moreover, exploring and visualizing data ecosystems require the
implementation of scalable methods able to traverse a variety of data sources. Finally, privacy
and data access control techniques are demanded to enforce regulations imposed by data
providers or data protection authorities. The project aims at developing data management
techniques that enable for the transformation of unstructured into a unified knowledge base.
Domain ontologies like UMLS, are used to annotate the extracted concepts; these annotations
provide the basis for entity matching and data integration. A unified schema is utilized to
describe the integrated entities and a federated query engine supports the exploration of the
knowledge base. Moreover, services for data analytics and prediction allow for the discovery
of novel patterns and associations that may explain the survival time and disease progression.
Initial results suggest that these knowledge discovery techniques on top of a knowledge base
have the power of uncovering relevant patterns in the integrated clinical data. [71, 111].
5.1.7 General Challenges in Health Applications
5.1.8 Multitude of Stakeholders and Patient-centric Data Exchange
In the health domain many different stakeholders with diverse interests, rights, and respons-
ibilities exist. In contrast to other domains, not only organizations are involved in use cases
for DE, but also single persons, such as patients and physicians. In many cases personal
data is involved which poses many challenges to the data ecosystem especially in terms of
data access and data control. The role of the data owner may be taken by a party (e.g., the
patient) different from the one who controls access to the data (e.g., the physician or clinic).
This opens up many questions as data owner and “data controller” need to clearly negotiate
which (transitive) rights the data controller may have and rules on how data exchange and
usage is handled need to be defined. Though the handling of personal data is quite strictly
and clearly regulated by laws, such as the GDPR, it is not an easy task to define a framework
which suits all possible use cases and stakeholders’ needs. A patient-centric data ecosystem
100 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
would need to enable transparency of the data exchanged and of the operations executed
on them, making data provenance and auditing come to the fore. A patient-centric DE
should give a patient the opportunity to control what is exchanged with whom and with
what constraints. Finally, it should convey trust, that data is handled correctly to the rules
and constraints negotiated in the DE. Solutions to implement access and usage control for
DE have already been proposed, e.g., in the International Data Space [
]. These differ in
various aspects, such as the invasiveness for the internal systems (which is a crucial point for
health domain use cases) or the policy languages they are using.
5.1.9 Data Anonymization vs. Data Usefulness
A further concern is the trade-off between possible anonymization of personal data and
its usefulness. Depending on the use case, anonymization of the data is a must, e.g., in
clinical trials and the analysis of the results, or it is desired by the providing party to
not disclose certain data. But the anonymization may reduce the usefulness of the data
for further analysis. Hence, it is interesting to analyze the trade-offs between the extent
of anonymization and the resulting usefulness. These can be used to check against the
requirements of an application intending to use the data. Usefulness and anonymity could
be criteria to search for suitable data sources in a data ecosystem.
5.1.10 Data Variety, Standards, and Interoperability
In the health domain, a huge variety of data formats is used, many of which do not apply
recognized standards, but rather are proprietary to specific devices or systems. Although a
multitude of standards exist in the health domain, these are not used consistently and for
many crucial concepts, such as Electronic Health Records, a general standard is still missing.
In these cases a DE has to demand a precise documentation and meta data description of
the data provided to make the data interoperable for other parties. In the first (professional
health services) and second health market (consumer health and well-being services) many
insulated applications and data silos exist. Applications are just written for a specific purpose
or device, but are not able to connect to other systems or even export into a standard format.
To make this data reusable in other contexts, other services need to encapsulate it or wrap
it. Hence, a DE should provide means to easily integrate these data sources into a DE, e.g.,
enable services, which transform the data into an accepted standard or (semi)-automatically
add semantic annotations. For these services extra costs for data consumers or providers
could be charged.
5.1.11 Meta-data Usage, Quality, and Mappings
Many medical taxonomies and ontologies exist to express meta-data and to annotate data.
This is beneficial on the one hand to make the data more “understandable” and enables more
intelligent applications, which also consider the context and meaning of the data. On the
other hand, many of the most used vocabularies are so huge, such as MeSH, SNOMED, or
the NHI Thesaurus, that working with them and maintaining them is very difficult. This
results in quality issues in the vocabularies and, as a consequence, also in the applications
that are based on them or the mappings created between two or more ontologies. Also data
curation as such is an error-prone and tedious process, leading to poor annotations when
this is performed by untrained persons or by algorithms.
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 101
5.1.12 Data Quality
Data quality (DQ) plays a crucial role in all data-intensive applications, hence, also in
medical applications. There exist several sources of DQ problems in health applications.
Much of the health data is recorded or transformed using electronic devices and sensors
which may have an inherent physical imprecision or are easily prone to failure and lead to
DQ problems. The monitoring of DQ for online or streaming sources is difficult and may lead
to additional problems, such as synchronization and timeliness problems. Additionally, a lot
of data is collected manually, e.g., using paper forms, which are also prone to errors, missing
data, or transfer errors when digitizing the data. Furthermore, a multitude of data sources
and formats may be involved in just one application. The conversion between formats, the
integration with other data into existing data sets or databases, and the aggregation of data
may as well lead to DQ problems. Many standards, for example FHIR, provide guidance to
structure the data accordingly, but it can be very different how the guidance is implemented
which may also lead to problems in the data format and interpretation. Also, the semantic
consistency of data (is the data representing the patient and her condition correctly?) of
single and between multiple examinations and measurements must be checked which often
has to be a manual task. It is interesting to investigate how integration of several sources
and Machine Learning algorithms could be used to support this process.
5.1.13 Data Protection and Data Sovereignty
Working with personal data is highly regulated by laws on different levels. For example,
considering Europe and the EU, data security and protection is regulated by the General
Data Protection Law (GDPR). Each country in the EU additionally may have a national law
(e.g., in Germany the Federal Data Protection Law), which extends the directly applicable
GDPR. Further regulations can be defined on state level (e.g., the Krankenhausgesetz in
Bavaria or the Data protection Law in North-Rhine Westphalia). These multiple overlapping
regulations lead to a highly complex environment for data ecosystems and data exchange.
In turn, this requires a highly flexible framework which must combine usage control, access
control, data provenance, and further means to enforce data protection and enable data
Data sovereignty for patients is desirable, but not easy to implement. Patients may not
feel capable of deciding which information should be accessible to whom. It is hard to decide
for laymen, what exactly is sensitive data and if this data is crucial to a specific health
professional to enable the best possible treatment. The grade of transparency regarding
data exchange may also be subject to discussion, as for example a very “chatty” notification
protocol may annoy or make patients feel not secure.
5.1.14 Conclusion
In this section we have presented two use cases which constitute complex DE in different
ways, but sharing similar problems. In general, the challenges for data exchange are manifold
and specific in the health domain as many stakeholders with many different interests are
involved. Especially, single persons, such as patients and doctors, play a major role in the
DEs which requires to overthink mechanisms so far identified and implemented for industry
and companies. Highly sensitive data and the corresponding ethics and laws around it,
pose crucial challenges to DEs. And finally, the high variety in data formats and metadata
definitions make the domain description and application implementation special. Hence, we
postulate that a framework for DEs has to take all of these challenges and requirements for
the health domain into account to propose general solutions.
102 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
5.2 Industrie 4.0 Data Ecosystems Examples
Egbert-Jan Sol (TNO – Eindhoven, NL)
License Creative Commons BY 3.0 Unported license
©Egbert-Jan Sol
Multi-sided markets are based upon an alliance agreement for sovereign data exchange
between organizations and result in a distributed data driven ecosystem. In such data
ecosystems (DES) each organization can do more then when using only its own data, and
once properly architectured this can be realized without the need to see all data such that
each party remains control over its own business.
Industrial data ecosystems are encountered in the (discrete) manufacturing, the (chemical)
process industry, in (maintenance) and logistics services, but also in amongst other industries
as telecommunication ecosystems as GSM. In practice all these industrial cases there is a
physical device that can get a so-called digital twin, being the digital representation of a
physical good. With this digital twin more data can be linked to the device resulting in data
sets that can be (partly) shared. In industry we encounter data sets that consist of sensor
data, copyrighted data as drawings all the way up to very valuable, confidential data sets.
We introduce in this section four notions: 1-a data analogy of material, 2-digital twins
and identifiers, 3-four classes of data and finally 4- a function of controlled data exchange. It
is followed by four industrial use cases of data ecosystem: GSM mobile telephony, discrete
manufacturing, traceability of food/goods, and preventive maintenance with hints to aspects
as use of an association, international standard and cyber security protection consequences.
5.2.1 From data element to data sets with sovereignty control to value
Data records, due to their digital nature, are often seen as discrete elements. But to accept
that certain data is more valuable than other data, an analogy with the material/process
industry might be more applicable.
Say, an individual sensor reading could be seen as atom and a long set of sensor readings
as base material. Sensor data as such is not so valuable and sensor data is not copyright
protected. Then if one combines iron with carbon to produce steel or mix a polymer with
a color ingredient one can get a stronger alloy due to the iron/carbon matrix/lattice or a
colored plastic. Combine two data set also leads to a more valuable set then just a single
list of data points, just as steel is more valuable than its base materials. Next image the
combination of a drawing and a set of manufactured parts that are verified to be produced
from the steel alloy or plastic within specs according to the drawing. This set is more valuable
than just a bag of polymer granules or a block of steel. Data sets that are verified are also
more valuable.
We won’t identify the single sensor data element as we won’t identify an individual atom.
The base material is already more valuable, but not as valuable as a strong alloy. Similarly
we might identify a list of sensor points having some value, but the combination with a
safe boundary set within the sensor points must stay is already more valuable. And an end
product similar as a list of customer bills is even more valuable.
There is an analogy in data set similar to atoms, to base materials to chemicals, to discrete
products up till a product owned by someone. Single data lists, data sets and their relations
and finally a whole data bases can be have a certain value for a user. And combining data
sets over multiple parties and share it in a larger data ecosystem can even be more valuable.
The challenge is to construct the data ecosystem such that parties can do more with the
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 103
Figure 9 Nested Digital Twin with data elements.
available data without the need to see all data. Similar as you driving a car without knowing
and having the access to all the internals, the manufacturing secrets as costs of components.
But whereas a product has an owner, a manufacturer, a user, etc., data can be copied
over and over again. Here data ownership is more difficult to define. Only copyrighted data,
i.e. where human creativity is involved as e.g. drawings, can have the copyright owned by a
legal entity. And for certain data sets there exists a databank act. For private data Europe
has the notion of data sovereignty in the form of the GDPR. As of today, in industry there is
no legal binding concept a data ownership and/or sovereignty. There only exist contractual
relations on the sharing and use of data between parties. Only the producer of the data has
sovereignty control over whether to share the data with others. But e.g. sensor data has no
copyrights and once shared without a contract other can do what ever they want.
5.2.2 Industrial data twins and data identifiers
Each physical object can have a digital representation in the form of a data representation
as simple as just a number or as complex as a large data set with drawing, manufacturing
recipes and use history. In general, it will have a name and a digital identifier too. In figure
9 a physical object or asset with its digital twin as a kind of administrative shell around it is
An object can consists of subpart or a group of objects can form a larger object. This
results in a hierarchy of digital twins each with their own data and links to other digital twins.
In particular similar objects could each share the same design and their own instantiation.
In that case, these digital twin instantiations (DTI’s) share a common design (see figure 10
with the DT on the left and the DTI’s on the right).
The design part of a digital twin could be a set of drawings, software, recipes etc. This
design and the digital twin (of the design) (DT) is owned by a legal entity. And the physical
objects can be owned by (other) legal entities. But what about ownership of the digital twin
instantiation of each object with e.g. the historical (sensor and use) data?
5.2.3 Confidential, shareable, shared and public data sets
Each digital twin of an object, a system or even an organization can have confidential
information that is not (to be) shared by anyone else. The owner of an object can also decide
that certain data can be shared with other legal entities with whom the owner decides to
104 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
Figure 10 Digital Twin with its DT and DTI (instance).
Figure 11 Anatomy of a Data EcoSystem.
share data with. This requires both a technological standard to exchange the data as well
as a legal contract to specify the conditions of sharing. One can also receive data made
shareable by one or more other parties, in a similar way the owners made some data shareable
to others, again using tech standards and legal contracts). Finally, there might be public
data that can be used too, see also figure 104 on the anatomy of three different kinds of data
ecosystems illustrated as isolated islands, winner-takes-all and shared common. The last one
is the preferred one in case one wants to keep a certain data sovereignty.
The result is that around every physical object, systems and even organizations a digital
twin can exist with data and that these digital twins can form a data ecosystem. In industrial
cases that data ecosystem will be related to physical objects, digital identifiers, data sets and
relations between data sets where the digital twins can exist and be copied unlimited times
in all kind of subsets while there is only one instant of the object.
Different players in a data ecosystem can have different views of the data and therefore
different digital twins of the same object. Of course, by expanding the own view of the digital
twin data with data shared by others a better and often more valuable digital twin can be
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 105
Figure 12 Different Data Ecosystem architectures.
To function in a data ecosystem, it is therefore sensible to make some or all of the own
data on that DT object and/or DT instances shareable to others too. Depending on the
business parties can agree to keep as much confidential (not shared) and minimize shareable
data. In an extreme case one party could try to collect as much data from the others and
maneuver itself in a data monopoly position. It depends on the parties how much they are
willing to share, the required contracts and the technical interfaces they select and use for
data exchange.
5.2.4 A Data Ecosystem function
To be able to share shareable data between parties in an ecosystem one needs, next to
technical data-communication standards and legal contracts a software function. With X the
data and C the legal constraint put on the data exchange, we call the entity that controls
the exchange of data T (transmit). And with Ps (or S) the sending party and Pr (or R) the
receiving party, T performs a function f(X, C(S, R)). T receives data X from an Ps under
specified constraint instruction. See figure 12.
T can be distributed as in e.g. IDS over all parties or T can be a clearing house with
processes the data as specified. Any party P could have a sending and a receiving side. In a
data ecosystem there can be many S’s and R’s. An unbalanced situation is when there are
many S and only one R, in particular if T is completely under control of the legal entity R.
A tightly coupled chain might be when a flow existing from S to R follows by R to O, O to
P, etc. In an hierarchical system many parties send and receive data up- and downwards and
not so much side wards. Finally in an fair data-ecosystem all parties can exchange data with
everybody else.
5.2.5 The GSM model
The GSM association runs a data ecosystem since the 1990-ties. With mobile telephony 1
users could not roam. With GSM a user can roam to a geographical different network of
another GSM operator as in cross border travel. Because of the roaming agreement a GSM
operator can provide more services then in generation 1 mobile telephony networks. But
GSM operators need to share data on the roaming users. This is done by the concept of home
and visitor location centers. In essence this is data-ecosystem that provides its participants
more services then without it. See figure 106
In the home location center the user, the GSM nr (and device ID (IMEI)), the used
service and the billing information is known. In the visitor location center only the device
ID with the usage of mobile voice time and data traffic is known. This information is shared
with the home GSM who then pays a bulk fee to the visited GSM operator and adds up the
received usage data to the customer usage to finally produce a bill. Next to the realtime
106 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
Figure 13 GSM network.
exchange of visiting or roaming data of the device there is the contractually data exchange
on the total usage of visiting devices. Notice that te home GSM operators has personal data
as customer name, billing info and GSM nr. The visiting GSM operator only has the device
ID. When ever a device roams it device ID indicate the home GSM operator and visiting
operator informs the home location register in which visitor location the device currently
is. When the mobile phone number is called (in the home location visitor), the GSM home
operator routes the call to the visiting GSM operator.
In the GSM case the identifiers are the home/visitor location centres, the GSM device
ID each linked data sets as customers, device usage, etc. The transport T has three simple
functions: updating the device visitor location (in realtime) to the home, forwarding calls to
the device in the visiting network and sending the usage at aggregated level to the me too.
5.2.6 Discrete manufacturing supply chain
This example is a first-tier supplier network where orders from the OEM-er are sent to a
supplier including o.a. drawing information and where finished subparts are sent to the
OEM to built the product. Without the order and, if it is not a standard component, the
drawing information, the supplier cannot produce. Often parties try to keep as much data as
possible confidential, but more and more data is sent to the OEM-er, often on demand of the
OEM-er for traceability, quality monitoring and control of the (realtime) logistics in the chain.
Sometimes an independent player T could be used to assemble business information as market
share and relative quality information enabling suppliers to upgrade their performance. In
other cases the OEM can give feed back information to the supplier performance.
As with the GSM example, here to digital identifiers, but also digital twins and a three
level physical object, digital data exchange and legal contracts are encountered.
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 107
Figure 14 Supplier - Original Equipment Manufacturer network.
5.2.7 Food traceability
More and more industrial chains demand extensive traceability. Examples are the automotive
and aircraft industries, but also the food industry. In figure 1 a warehouse with sensors was
shown. In that case it is a part of chain where food is stored under certain temperature
and humidity conditions. The warehouse operator uses the data to optimize the energy
consumption, but the same data can also be shared or “sold” to the food owner to monitor
the storage conditions for traceability and quality control. But sharing the data, with or
without payments, the whole (data)eco system can perform better.
5.2.8 Predictive Maintenance
Predictive Maintenance is similar to the food traceability use case. In this case a maintenance
service party monitors the condition of equiment to predict a stop for maintenance. Often
the operator of the system uses the data for (real-time) control at the same time. By sharing
the data with another party that party can combine more data for others to improve its
prediction algorithms. The specifics of this use case is that these environment tend to require
strict security and access control.
Figure 15 shows the case where OT (operational technology) has shielded data access
in its own subnet and data cannot be sent directly to others. In this case the data is sent
by OPC-UA (IEC standard) protocol through a firewall to a server to be stored there and
made suitable to be further processes outside the operational production environment. These
two level data ecosystems often restrict other legal parties to have an active proces directly
communicating through the customers network to their own remote services support networks.
OPC-UA is used for equipment communication within an organization. Other standards
as the IDS (international data spaces) can be used for data exchange between different
organization using the IDS technical and contractual agreements.
108 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
Figure 15 data ecosystem inside and outside a factory.
5.3 Use Cases from the Smart Cities Domain
Cinzia Cappiello (Polytechnic University of Milan), Bernadette Farias Lóscio (Federal
University of Pernambuco), Avigdor Gal (Technion – Haifa, IL), Fritz Henglein (Univ. of
Copenhagen, DK &Deon Digital – Zürich, CH)
License Creative Commons BY 3.0 Unported license
©Cinzia Cappiello, Bernadette Farias Lóscio, Avigdor Gal, Fritz Henglein
The Smart city concept has various definitions that are associated with viewpoints that
range from the people perspective to a more technological perspective. Such latter point of
view suggests that smart cities mainly focus on adopting the next-generation information
technology to “all walks of life, embedding sensors and equipment to hospitals, power grids,
railways, bridges, tunnels, roads, buildings, water systems, dams, oil and gas pipelines and
other objects in every corner of the world, and forming the “Internet of Things” via the
]. Smart Cities data collected by heterogeneous devices and data sources are
strategic to perform analysis, to extract relevant patterns, to detect inefficiencies, and to
propose innovative solutions to improve the quality of life of citizens.
5.3.1 Scenario description
Goal: Definition of an ecosystem that includes several cities (learning from one city can help
to accelerate the process in another city).
The Smart Cities DE, presented in Figure 16, can be seen as a composition of Data
Ecosystems. Each city has its own DE and they can exchange data and knowledge. Each DE
is composed by Data Providers and Services. The Data Ecosystem of City 1, for example, is
composed by five Data Providers, which can also play the role of a Data Consumer. Data
from different providers can be used by a Data Consumer to offer more advanced service.
In this scenario of data mobility, Data Provider 1 and Data Provider 3 provide bus data
to Data Provider 2, which will perform time predictions based on these data. In a similar
way, Data Provider 4 receives bus data from Data Provider 1 as well as time predictions
from Data Provider 2. Data Provider 4 uses these data to develop a better location planning
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 109
Figure 16 Smart cities data ecosystems.
for bikes to rent. Finally, Data Provider 5 uses the services offered by the DE to provide
information to citizens.
In such ecosystem, we have to clarify that a city can learn from another city but it is
difficult that a city might inherit or use the same applications/services of another city. In
fact, most of the smart city applications are context- aware and tailored for the city needs.
5.3.2 Challenges
General challenges
It is important to guarantee that people use data in the correct way: data usage
constraints have to be defined.
Looking at the data infrastructure, it is difficult to understand which are the boundaries
of a node. It might be adopted a data provider-oriented approach in which a node is a
data provider within its own data sources.
Challenges related to derived data issues
In the data ecosystem of a city, in general a data provider can be the owner of raw
data sources or can provide derived data created transforming data gathered by other
data providers. In summary we can distinguish between raw data and derived data. It
is important to find a way to describe and manage derived data.
How do we define the usage constraints of the derived data? It might happen that raw
data are regulated by usage constraints but derived data do not contain personal data
and vice-versa. Besides, integration between two sources could reveal personal data.
110 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
The code related to the applications/services should be represented. Also the code
should be protected
View maintenance: model improvement or data improvement.
Challenges related to Quality assessment
In this architecture, exchanged data should be high quality data in order to exploit the
potential data value. It is necessary to define who should be in charge of the quality
assessment. In fact, data quality assessment could be centralized and in charge of the
platform through a platform service also using crowdsourcing.
6 Manifesto
The Dagstuhl Manifesto is an agreed result of the Dagstuhl Seminar “Data Ecosystems:
Sovereign Data Exchange among Organizations” that took place on 23-27 September 2019 in
the Schloss Dagstuhl Leibnitz Zentrum fur Informatik, that joined forces of the researchers,
practitioners and experts from Europe and worldwide, with the goal to discuss new challenges
in building data ecosystems supporting sovereign data exchange among organisations involved
into whole data value chain. We understand that:
Data exchange among organizations is a key enabler for the digital economy of the future.
A secure, reliable and performant data exchange infrastructure is a basis for operating
sustainable data ecosystems supporting the whole data value network involving data
providers, data consumers, and service providers.
Enabling assessment and awareness of data quality are core requirements in a data
Organizations and individuals must be regarded as data sovereigns entering into data
ecosystems according to agreed contracts.
In a data ecosystem, data should be considered both an economic asset and a tradable
commodity according to specified conditions of use.
Defining metadata is necessary for enabling a variety of operations on data and with
data, along the whole data value network.
Defining semantic interrelationships among data sets is a key problem/task in creating
scalable data ecosystems supporting effective data exchange.
Defining a generic data ecosystem architecture is necessary for supporting interoperability
at multiple ecosystem levels, including data sources, semantics, applications, workflows
and processes, governance and economics.
Economically viable data ecosystems architectures should leverage successful experiences
from similar distributed systems such as GSM employing context roaming between
independent domains.
Future data ecosystems should serve the public interests, in particular by supporting
data related projects to address the UNDP 17 Sustainable Development Goals (SDG) .
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 111
7 Statements of the participants
7.1 Cinzia Cappiello (Polytechnic University of Milan, IT)
My research focuses on Data and Information Quality. In particular, in the last years, I
addressed issues related to the Data and Information Quality assessment and improvement in
the context of service-based applications, Web applications, Big Data and IoT. Data Quality
is defined as fitness for (intended) use, that can be seen as the capability of a data set to
satisfy users’ requirements [
]. This definition suggests that quality is subjective: a data set
that is appropriate for an application/user might not be suitable for another one. For this
reason, currently, I am working on the design of an application-aware data quality assessment
platform: data quality should be evaluated by considering the actual usage of the considered
data set. Moreover, in this field, most of the literature contributions propose approaches
for structured data sets. In data ecosystem, the increasing volume and variety of the data
sources needs the definition of new data quality methods. Such methods should be designed
by considering the type of source (e.g., data streams vs traditional db) and the type of data
(e.g., text vs numerical). In fact, these two variables impact on the dimensions to evaluate
and on the metrics to use. In summary, in order to enable an automatic evaluation of the
quality level of the data sources it is necessary to define an adaptive data quality assessment
service able to trigger the appropriate mechanisms by considering the characteristics of the
sources and the application/user that requested data.
7.2 Ugo de’Liguoro (University of Turin, IT)
My interests are to logic and computation. Central are both pure and typed
-calculus. I
have also investigated other formalisms, like the
-calculus and the
-calculus, considered
as theoretical calculi modeling object-oriented and concurrent programming languages re-
spectively. Key methodologies in my work are type systems and denotational semantics, that
nicely correspond each other especially in the case of intersection type assignment systems.
The latter express functional properties of
-terms seen as programs, and I have worked to
include in the theory also non functional aspects, like non determinism and control operators.
I believe that intersection types have a great potential to model both behaviour and data;
an experience in that direction is the participation to a project of program synthesis from
components in Dortmund, based on intersection types and combinatory logic. In a related
field I have been working on session types and contract theory. This is about modeling
protocol compliance of multiple principals interacting through a network, enjoying safety
and liveness properties. A particular concern in this work is checking for consistency of local
protocols to form a well behaved system, and the adaptation of each to the others when
considered as components independently built and specified.
7.3 Yuri Demchenko (University of Amsterdam, NL)
The fact that data has a value is commonly recognised. However, data value is different
from those associated with the consumable goods. There are a number of initiatives to
create data markets and data exchange services. Well established business models of paid or
commercial data(sets) services such as data archives are based on the service subscription
fee. Quality of datasets in many cases is often assessed by independent certification body or
112 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
based on peer review by expert. However this model does not provide a basis for making data
an economic goods and enable data commoditisation. Another important development in
making the best use of research data is based on wide implementation of the FAIR (Findable
– Accessible – Interoperable – Reusable) data principles, which are widely supported by
research and industry. However, emerging data driven technologies and economy facilitate
interest to making data a new economic value (data commoditisation) and consequently
identification of the data properties as economic goods. The following properties proposed in
] leverage the FAIR principles and are defined as STREAM for industrial and commoditised
data: [S] Sovereign - [T] Trusted - [R] Reusable - [E] Exchangeable - [A] Actionable - [M]
Measurable Other properties to be considered and necessary for defining workable business
and operational models: nonrival nature of data, data ownership, data quality, measurable
use of data, privacy, integrity, and provenance. Defintion of the data properties as economic
goods must be supported by creating consistent and workable models for data exchange and
commoditisation, to facilitate creation of new value added data driven services. Consistent
data pricing and data markets models are equally important for government funded and
sponsored research, open data and governmental data. The proposed Open Data Market
(ODM) model is based on the adoption of the IDSA Architecture and data sovereignty
principle [
]. The ODM must be based on relevant industry standards and provide secure
and trusted data exchange between data market actors: data producers/owners and data
consum-ers, services and applications developers and operators. A functional data market
model and architecture should include multiple components such as the secure trusted data
market infrastructure as well as regulatory basis. The proposed open data market model
would be decentralized and allow creating virtual private market instances to support data
exchange between peers or group of peers. This gives the advantage that network nodes,
data sellers and data buyers, receive the full benefits of the data market while retaining
full control of their data. This is an important requirement for data markets for industrial
data where companies, the data owners, want to retain control over their data, maintaining
data sovereignty. The ODM research evaluates the blockchain technology to enable an open
controllable trusted data market environment for secure and trusted data exchange, and
support data value chain (provenance) and create a bias for data monetisation 13 .
7.4 Elena Demidova (Leibniz Universität Hannover, DE)
Recently, we have been involved in several projects focused on analysing data for urban
mobility use cases. Examples include prediction of impact of planned special events on traffic
], identification of structural dependencies in urban road networks [
], or analysis
of driver behaviour. Realisation of such use cases requires integrated analytics of data
originating from different domains (map data, traffic information, car trajectories, car sensor
data, city infrastructure data, population statistics, people movement, etc.), as well as
different sources and owners. Currently, most of the data relevant for such analytics is locked
by different organisations (e.g. companies or city municipalities), making it hard to obtain
the overall picture and build accurate prediction models. In this seminar, I would like to share
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 113
experiences we have collected in several related projects (including Data4UrbanMobility
], Simple-ML
] and CampaNeo
, leading to challenges related to collecting such
data, and making use of the data in the analytical use cases more transparent to data owners.
7.5 Boris Düdder (University of Copenhagen, DK)
Boris Düdder’s core interests are in formal methods of software engineering and distributed
systems, for example, by employing logical methods, rewriting systems, and model checking.
His interests concerning the Dagstuhl Seminar are methods for automatic configuration
of exchange of sovereign production data of manufacturers and suppliers securely while
guaranteeing ownership, confidentiality, and privacy. He can contribute his expertise and
industrial experience in formal methods for synthesis, analysis, and verification of software
for industrial production systems and supply chain (including logistics) as well as distributed
ledger technology, e.g., for proof of provenance applications and finance. Industry experience
includes logistics, whiteware and aircraft manufacturers.
7.6 Bernadette Farias Lóscio (Federal University of Pernambuco, BR)
In the last years, I have been involved in several projects related to Open Data and Data
on the Web. One of my main interests concerns how to share data on the Web in a proper
and sustainable way. The growing interest in sharing data on the Web gives raise to several
challenges related to important subjects including data provenance, data privacy and data
access. In order to help data providers and data consumers to face those challenges, the
W3C Data on the Web Best Practices Working Group proposed a set of 35 Best Practices,
which cover different aspects related to data publishing and consumption, like data formats,
data access, data identifiers and metadata. I was part of the DWBP working group and I am
one of the editors of the Data on the Web Best Practices recommendation. This experience
was very rich and I could have the opportunity to identify several open research challenges
related to Data on the Web. One of these challenges concerns the creation of sustainable
Data Ecosystems.
While Data Ecosystems are gaining importance, several ecosystems are not sustainable
and consequently the effort spent by their actors end up not being properly used or forgotten.
The lack of communication and cooperation between data producers and consumers is one
of the main obstacles moving towards sustainable Data Ecosystems. Moreover, designing,
developing and further maintaining systems for Data Ecosystems are not trivial. Recently, I
have advised several students in subjects related to Data Ecosystems, including a metamodel
to represent data ecosystems, a metadata curation framework for data ecosystems and a
framework to assess data ecosystems health. In this seminar, I would be glad to share
my experiences on these subjects and I am willing to learn from the experiences of other
114 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
7.7 Avigdor Gal (Technion – Haifa, IL)
My research focuses on effective methods of integrating data from multiple and diverse
sources, in the presence of uncertainty. My current work zeroes in on schema matching
– the task of providing communication between databases, entity resolution – the task of
identifying data elements that relate to the same real-work entity, and process matching –
the task of aligning process activities. In the context of this seminar, I have contributed to
the design of an intelligent data lake, one in which integration is seamlessly generated and
tested with the assistance of experts, heuristics and machine learning algorithms
7.8 Sandra Geisler (Fraunhofer FIT – Sankt Augustin, DE)
My field of expertise and research comprises several aspects of data management. Especially,
my work is concerned with techniques in the area of data stream management, big data in
general, data quality, data lakes, meta-data management, data integration, and semantic web
technologies. Furthermore, I have a strong background in the medical domain and medical
Currently, I am working on projects in the context of the Fraunhofer Medical Data
. The Medical Data Space can be viewed as a distributed data ecosystem which
has specific requirements stemming from the challenges of working with health data. It
is especially concerned with data sovereignty, data security, and data privacy striving for
compliance to data protection laws on various levels. The data in the Medical Data Space
remains in custody of the data owner and is only exchanged in a controlled and secure manner.
For example, data important for the follow-up care of patients after a surgery may be shared
between several parties involved in the care process, such as the general practitioner, the
outpatient nursing service, the hospital, and of course the patient herself. But, every party
may only need a specific part of the data, or should not be allowed to access all of the data,
or should only see results produced by algorithms using the data as input. Furthermore,
data quality plays a crucial role as the data may be used to support processes which may
have indirect or direct impact on the health of a person. Also data sources with potentially
erroneous data, such as health monitoring devices, are often involved. But, controlled and
automated data collection of the users may also help to elevate DQ which in consequence will
also lead to a higher data value and quantity for health care services or other data consumers,
such as research studies. Finally, challenges regarding the integration and interpretation of
the distributed and very heterogeneous data have to be tackled to provide a use case specific
and user-friendly view on the data.
Based on my research interests and current project work, I can contribute to the seminar
in terms of the discussion of applications from the health care sector and their specific
challenges and requirements focusing on data sovereignty, data quality, and data integration.
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 115
7.9 Benjamin Heitmann (Fraunhofer FIT – Aachen, DE & RWTH
Aachen, DE)
The data market places of the future will be enabled by technologies which protect the value
of data for all participants of a market place. In my experience, too often the real value of
data or other digital assets for a stakeholder is neglected, as processes and algorithms can
currently be better protected. In order to protect the value of data in a market place, the
three goals of security, privacy and data sovereignty need to be fulfilled.
means protecting the digital assets which an organisation controls and which
might be valuable to an attacker. Examples of assets which usually are secured, include
domain-specific knowledge collections such as life-science data sets, financial records, records
of human resource data, product sales data, and instance data for business processes. Failure
to protect the digital assets of an organisation results in losing control of knowledge, processes
and customers, as competing organisations could gain access to the same assets.
means protecting the data of individual users, which is required as input for
many digital value chains. Examples of private assets include digital health records, purchase
histories for e-commerce sites, historical data about consumption habits for news, movies and
music. Failure to protect the personal data of individual users will result in unwillingness or
reduced incentives for users to share or sell their data. This will in turn disable any processes
which require input data from users to generate more value for an organisation.
Data sovereignty
means protecting the digital assets which an organisation or individual
controls and which have been sold or shared willingly with another entity. Examples of
assets which might be shared or sold without losing control include R&D data sets for
manufacturing, R&D data sets for discovering new gene interactions, personal data about
historic fitness activities or health indicators such as blood pressure. Failure to protect assets
after sharing or selling them will result in loss of control over the purpose of using the data,
and in decreased incentives to share or sell the data with additional entities in the future.
In future data market places, the data protection needs to fulfil the requirements of
all stake holders at the same time: data generator, seller and buyer. In addition, usually
all three goals of security, privacy and sovereignty need to be reached at the same time.
This requires technologies which enable
enforceable guarantees for protecting data
during processing
. In my experience, some of the technologies with the highest potential
come from new advances in information processing and encryption. The most notable new
developments are in the areas of secure multi-party computation (SMPC), homomorphic
encryption (HE), and differential privacy, anonymisation and synthetic data generation.
Together with new architectures, business models and legal frameworks, the listed tech-
nologies for enforceable guarantees to protect data during processing can enable sustainable
data markets which fulfill the requirements of all participants.
7.10 Fritz Henglein (Univ. of Copenhagen, DK & Deon Digital –
Zürich, CH)
Fritz Henglein is Professor of Programming Languages and Systems at DIKU, the Department
of Computer Science at the University of Copenhagen (UCPH) and Head of Research at
Deon Digital AG, a Zürich/Copenhagen-based start-up developing secure and scalable digital
contract technology for both decentralized (blockchain, distributed ledger) and centralized
systems. His research interests and contributions are in semantic, logical and algorithmic
aspects of programming languages, functional programming, domain-specific languages,
116 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
digital contracts, reporting and analytics, smart contracts and distributed ledger technology,
with applications in enterprise systems, business processes, high-performance computing,
probabilistic programming and decentralized systems, including blockchain and distributed
ledger systems.
Key desiderata of (next-generation) distributed ledger (DL) systems are tamper-proof,
privacy-respecting recording of real-world and business events together with their evidence;
secure distributed storage and transfer of assets such as money, assets, securities and (digital
twins of) physical resources; automatically managed and enforced contracts that provide an
effective and auditable basis for privacy preserving collaboration, analytics and planning; a
balance of availability, consistency and network failure tolerance tuned to specific use case
characteristics; and, most importantly, organizational decentralization to minimize the need
for and competitive advantage of a dominant and controlling platform provider. Such a DL
system facilitates tracking of digital resources, including valuable data, across organizations
as well as transfer (and revocation) of control over them. Ongoing research, development and
commercial application (at Deon Digital) of smart digital contract technology on existing
commercial distributed ledger systems indicates that this may facilitate effective cross-
organizational data exchange by facilitating unforgeable auditable proofs of authorized data
7.11 Matthias Jarke (RWTH Aachen University and Fraunhofer FIT,
Data-driven machine learning methods are typically most successful when they can rely on
very large and in some sense homogeneous training sets in areas where little prior scientific
knowledge exists. Production engineering, management, and usage satisfy few of these criteria
and therefore do not show many success stories, beyond narrowly defined specific issues in
specific contexts. In contrast, the last years have seen impressive successes in model-driven
materials and production engineering methods, these methods lack context and real-time
Our vision of an Internet of Production, pursued in an interdisciplinary DFG-funded
Excellence Cluster at RWTH Aachen University, addresses these shortcomings: Through
sophisticated heterogeneous data integration and controlled data sharing approaches, it
broadens the experience base of cross-organizational product and process data. At the
method level, it interleaves fast “reduced models” from different engineering fields, with
enhanced explainable machine learning techniques and model-driven re-engineering during
As a common conceptual modeling abstraction, we investigate Digital Shadows, a strongly
empowered variant of the well-known view concept from data management. The idea of
Digital Shadows dates back to Platon’s famour Cave Allegory but which he illustrated the
limitations of all human knowledge – we can always only see partial perspectives on the
world. Fifty years of data management research confirm that the growth of data has always
outpaced our ability to deal with them, and we expect it to stay this way. Therefore, we
see strong limitations for the currently fashionable Digital Twins when real-time and large
scope are relevant, and propose to circumscribe them by well-structured collections of Digital
Shadows. Besides enabling real-time monitoring and control at an abstract level, Digital
Shadows are also well-suited for bridging interdisciplinary boundaries and communication
problems between research and practice. Several initial experiments indicate the power of
this approach but also highlight many further research challenges.
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 117
7.12 Jan Jürjens (Universität Koblenz-Landau, DE)
AI uses scientific methods, processes, algorithms, and systems to gain knowledge and insights
into data that exists in a variety of formats. Characteristic of the area is on the one hand the
high significance that it has. Based on the results obtained, numerous important decisions are
made concerning the individual or society as a whole: diagnoses, therapies, credit decisions,
spatial planning, etc. On the other hand, AI is characterized by the iterative and empirical-
heuristic approach by which knowledge is extracted and decisions are derived. From the
point of view of the provider of the data on which the AI-based analysis is performed, it is
important that the Data Sovereignty is preserved in the context of the data analysis, which
is closely related to the following aspects:
Data security
Data privacy
Transparency and explainability
Fairness / non-discrimination
Unfortunately, it is a significant challenge to be able to demonstrate whether or not these
aspects of Data Sovereignty are satisfied in a given situation involving AI based analysis.
In fact, it is already a challenge to just describe “correct behavior” of an AI based system,
because its results are usually not predetermined and can only be obtained through the data
analysis process, so it is in general not clear what “correctness” means in this context. The
challenge is also that the behaviour of a self-learning algorithm depends on the training
data to which it is continuously being subjected and therefore cannot be determined by only
considering the algorithm itself. We thus need a verification approach which can be efficiently
parameterized over the possible effects of the self-learning process without having to re-do the
complete verification whenever the self-learning leads to a change in behaviour. The proposed
talk discusses these challenges and presents an approach that supports the analysis of Data
Sovereignty aspects in the context of AI-based systems based on the tool-based analysis of
software design models in UML, which supports change-based verification and can thus deal
with the different variants of algorithm behaviour arising from self-learning. The talk is based
on work done as part of University of Koblenz’ research priority programme “Engineering
Trustworthy Data-intensive Systems” as well as within the context of the “International Data
Spaces” initiative at Fraunhofer ISST.
7.13 Maurizio Lenzerini (Sapienza University of Rome, IT)
Data interoperability refers to the issue of accessing and processing data from multiple sources
in order to create more holistic and contextual information for improving data analysis, for
better decision-making, and for accountability purposes. In the era towards a data-driven
society, the notion of data interoperability is of paramount importance. Looking at the
reseach work in the last decades, several types of data interoperability scenarios emerged,
including the following.
In Data Integration, we have a multitude of information sources, and we want to access
them by means of a global schema, that somehow accomodates an integrated view of all
data at the sources [35, 75].
In Data Exchange, we have a source databas, and a target database, and we want to
move the data from the source to the target according to some specified criteria [5, 66].
118 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
In P2P Data Coordination, we have a network of information nodes (peers), and we want
to let them communicate to each other in order to exchange data or queries [20, 76].
In Ontology-Based Data Management (OBDM), we have a collection of data sources and
an ontol- ogy representing a semantic model of the domain of interest, and we want to
govern (i.e., query, update, monitoring, etc.) the data at the sources through the ontology,
rather than by interacting directly with the sources [29, 76].
A fundamental component of all the above data interoperability frameworks is the
mapping. Indeed, put in an abstract way, all the above scenarios are characterized by an
architecture constituted by various autonomous nodes (called databases, data sources, peers,
etc.) which hold information, and which are linked to other nodes by means of mappings. A
mapping is a statement specifying that some relationship exists between pieces of information
held by one node and pieces of information held by another node. Specifically, in Data
Integration the mappings relate the data sources to the global schema, in Data Exchange
they relate the source database to the target database, in P2P Coordination they relate
the various peers in the network, and in OBDM they relate the various data sources to the
ontology. In the last years, many papers investigate the notion of mapping, from various
points of view, and with different goals (see [
] and references therein). By looking at
these papers, one could argue that one of the most important role of mapping is to allow
reformulating queries expressed over a node into queries expressed over other mapped nodes.
Such reformulation task is crucial, for example, for answering queries expressed over the
global schema in a data integration system. Indeed, to compute the answer, the system
has to figure out which queries to ask to the data sources (where the real data are located),
and this is done by a step that we call direct rewriting: rewrite the query over the global
schema in terms of a query over the data sources. A similar task has been studied in the
other data interoperability scenarios. In OBDM, for instance, given a user queries expressed
over the ontology, the aim is to find a direct rewriting of the query, i.e., a query over the
source schema, that, once executed over the data, provides the user query answers that are
logically implied by the ontology and the mapping. While the notion of direct rewriting
has been the subject of many investigations in data interoperability in the last decades, in
this paper we aim at discuss also a new notion of rewriting, that we call inverse rewriting.
The importance of this new notion emerges when we consider the following task in the
OBDM scenario: Given a query q over the sources, find the query over the ontology that
characterizes q at best (independently from the current source database). Note that the
problem is reversed with respect to the one where the traditional (direct) rewriting is used:
here, we start with a source query, and we aim at deriving a corresponding query over the
ontology. Thus, we are dealing with a sort of reverse engineering problem, which is novel
in the investigation of data interoperabilty. We argue that this problem is relevant in a
plethora of application scenarios. For the sake of brevity, we mention only three of them.
(1) Following the ideas in [
], the notion of reverse rewriting can be used to provide the
semantics of open data and open APIs published by organizations, which is a crucial aspect
for unchaining all the potentials of open data. (2) Although the architecture of many modern
Information Systems is based on data services, that are abstractions of computation done on
data sources, it is often the case that the semantics of such computations is not well specified
or documented. Can we automatically produce a semantic characterization of a data service,
having an OBDA specification available? The idea is to exploit a new reasoning task over
the OBDA specification, that works as follows: we express the data service in terms of a
query over the sources, and we use the notion of reverse rewriting for deriving the query
over the ontology that best describes the data service, given the ontology and the mapping.
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 119
(3) It can be shown that the concept of reverse rewriting is also useful for a semantic-based
approach to source profiling [
], in particular for describing the structure and the content of
a data source in terms of the business vocabulary.
7.14 Wolfgang Maaß (Universität des Saarlandes – Saarbrücken, DE)
Intelligent services using Artificial Intelligence methods are tools for creating, modifying
and merging data products from different industries, in particular industrial production
systems (Industrie 4.0). We take an extended view on data products, which includes (a)
data as statements about a domain, (b) software, and (c) models including conceptual
models, machine learning models, and semantic models. In our research we develop AI
algorithms and architectures for distributed (IOT) environments. These technologies are
used to build domain-specific smart service systems. For example, Smart Dialog Services
are used to make data products and derive decision recommendations based on machine
learning mechanisms accessible to human experts (explainable AI and responsible AI). A
further focus is the investigation of means for the automatic evaluation of data products as
input for the financial reporting of digital assets. Our research has been conducted in various
fields, such as industrial manufacturing, health care/medicine and media. While the media
industry and medicine have been struggling with the challenges of digital transformation for
about 30 years, industrial production is at the beginning of understanding digital products
as an independent economic asset class.
7.15 Paolo Missier (Newcastle University, GB)
Data marketplaces are becoming ubiquitous, but for the most part, they assume (i) static
data, and (ii) a trusted environment where data trading takes place. Both these assumptions
introduce limitations in the potential of individuals to exchange their own personal data
with other parties. Data streams that originate from Internet of Things (IoT) devices, often
placed in people’s premises (house, vehicle, or own body using wearables), are increasingly
viewed as tradeable assets with value not only to the device owners, but also with resell value,
i.e., to third party buyers. We envision a marketplace for IoT data streams that can unlock
such potential value in a scalable way, by enabling any pairs of data providers and consumers
to engage in data exchange transactions without any prior assumption of mutual trust. We
have recently proposed one such next-generation marketplace model, where the requirement
for trust in data exchange is fulfilled using blockchain technology and specifically Smart
Contracts. I am interested in discussing such marketplace model within the context of this
Dagstuhl Seminar, as I believe it has the potential to become a cornerstone of data ecosystem
where organisations exchange their data reliably and in real-time, subject to legally binding
trading agreements, and while providing incentives for fair trading.
7.16 Boris Otto (Fraunhofer ISST – Dortmund, DE & TU Dortmund,
The proliferation of digital technologies such as cloud platforms and cyber-physical systems
in the manufacturing industry enables process and service innovation in production and
supply networks. Sharing data among network partners in an important prerequisite for
120 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
leveraing this innovation potential. In this context, the notion of “digital twins” has received
significant attention, both in the scientific and the practioners’ community. In general,
a digital twin represents real world objects (such as production equipment, components,
material etc.) along their lifecycle. It consists of descriptive data (e.g. the eCl@ss number of
a tool machine component) as well as event data (such as temperature, vibration data etc.
captured in the production process). Furthermore, the digital twin comprises both metadata
and data items. A prominent digital twin information model, namely the so-called Asset
Administraive Shell proposed by the German “Plattform Industrie 4.0” distinguihes between
“types” and “instances”. Recent digital twin scenarios are based sharing digital twin data
within a production and supply network or even within an industrial ecosystem. Hence,
the concept of “shared digital twins” of components, machines etc. emerges. While the
fundamental concepts of digital twins in general (such as metadata, data models, distributed
data storage are mature research topics, sharing digital twin data among different partners
has not sufficentily be conceptualized. Moreover, taking an Information System Research
perspective, a set of interesting research questions can be identified. Examples are:
What are appropriate query approaches to retrieve shared digital twin data?
How to distribute storing and processing of shared digital twin data between edge, fog,
and cloud level? What are appropriate synchronization approaches?
How do ensure data sovereignty of different contributers to the shared digital twin in a
complex production and supply network?
How to design functional data governance models for shared digital twin?
How to enable access and usage control for shared digital twin data?
How to design data provenance architectures for shared digital twins.
Based on a solid conceptulization of shared digital twins, a research agenda is required to
identify research demand and outline promising research trajectories.
7.17 Elda Paja (IT University of Copenhagen, DK)
My research focuses on techniques to support the elicitation or design of socio-technical
requirements (looking not only at software, but also humans and organizations, their inter-
actions and message exchanges), requirements that come from different stakeholders, and
might be conflicting with one another or impact business policies or system functionality. In
my work, I have been exploring the design of socio-technical systems from different angles:
security, privacy, risk, and decision-making support. The common ground of my approach is
the use of conceptual modelling techniques, and in particular the use of modelling primitives
to describe the rationale behind the behaviour of the various participants of a socio-technical
system, and the definition and exploitation of automated reasoning techniques to support
the work of requirements engineers. Contributions to the seminar
Discussing on analogies and potential of adapting existing methods for requirements
engineering for data ecosystems, treating data as first class citizens.
Presenting recent work on consent privacy requirements for sociotechnical systems,
following a multi-level approach, starting from social and organizational requirements, to
business process level verification of deviations or breaches.
Presenting recent work on consent verification monitoring, namely a formal framework to
support companies and users in their understanding of policies evolution under consent
regime that supports both retroactive and non-retroactive consent and consent revocation,
all in a context where personal data provides important business value, e.g. in the
personalization of services.
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 121
7.18 Barbara Pernici (Polytechnic University of Milan, IT)
Sharing of scientific and scholarly data is enabled by open or shared repositories in many
different scientific domains. Data sharing and open data are not final goals in themselves,
however, and the real benefit is in data reuse by different actors.
Focusing on reuse, the design of integrated frameworks that make it simple and effective
both the upload and the retrieval of large amounts of scientific data extracted from the
literature or produced by research labs, to support research, and in particular scientific
model development, is a challenging issue. In particular the development of scientific models
to reproduce and predict complex phenomena is a challenging task, which requires a rich
set of data both for model development and validation. This task is becoming particularly
critical as more and more approaches to automate model development using machine learning
techniques emerge in different fields.
Starting from ongoing researches in the domain of chemistry engineering and emergency
information systems, several research questions emerge and will be discussed:
How to represent metadata in a flexible way, in particular when there is no agreement
within the ecosystem of contributors on a common underlying model and new features
emerge during the analysis.
How to allow several organizations simultaneously analyze the data from different per-
spectives and integrate their results.
How to assess the quality of the data and of the analysis/classification models, possibly
automatically derived through deep learning mechanisms, distinguishing between assess-
ment of quality of data (both training and input data) and other quality issues introduced
by the analysis models.
Another open issue concerns the intellectual property and access to results both of the
original data and of data obtained through modeling/ simulation processes.
Finally, the interaction of information systems experts, computer scientists and experts
on the specific domain of interest requires new flexible approaches and design patterns in
the development and of such systems.
7.19 Frank Piller (RWTH Aachen, DE)
One of our core research streams at the Institute for Technology and Innovation Management
at RWTH Aachen University investigates the need of established corporations to deal with
disruptive business model innovation and supporting organizational structures and cultures.
The rise of platform-based business models (or business models for industrial data ecosystems)
is one of the main drivers of change in this field. Hence, we currently have a number of
research projects where we study the systematic development of these platform-based business
ecosystems and the strategic positioning of an industrial company in these ecosystems. The
largest among these projects is the Cluster of Excellence “Internet of Production (IoP)”,
funded by the German Research Council (DFG), Project ID 390 621 612). The IoP resembles
the vision of an open network of sensors, assets, products, and actors that continuously
generate data. A core element hence is the (re-)use of data, digital shadows, and applications
by other parties to facilitate faster and more efficient learning and analytics. The rise of
platforms (business ecosystems) where these data are being exchanged and enhanced by
dedicated “apps” is a central element of the IoP vision. To create value, ecosystems build on
complementary inputs made by loosely interconnected, yet independent stakeholders. Among
122 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
these participants, dedicated mechanisms governing data access and privacy are required.
Our research here takes an inter-firm (external) perspective: setting the right incentives
for sharing deep production know-how and data while balancing value creation with value
capture (sharing the rents) for all participants. Among others, we are interested in the
following research questions:
Modeling the tension between openness in value creation and control of value capture,
Managing property rights (access, transfer, enforcement) at data, applications, and
connected assets as a result of varying degrees of platform openness, and
Definition of governance modes and design factors to generate adequate business models
for the IoP that allow to maximize value appropriation for all involved actors.
7.20 Andreas Rausch (TU Clausthal, DE)
Data today plays a much bigger role than it used and now data is the new oil. The reason
why data is generated and collected in such a massive amount is pretty simple. Data is the
new oil and has become a fuel for new business models. However, most of the available
data is related to very limited number of companies and organizations that form almost an
oligopoly – the so-called GAFA companies
18 19
: Google, Facebook, Apple and Amazon. For
Small and Medium Enterprises (SMEs) it is much harder to extract the same value from
their data compared to these large enterprises. On one hand it is very difficult for enterprises
(Data consumers) to obtain proper data and on other hand for those who collect data (Data
providers), the problem is- How to draw to additional profit from the data beyond its obvious
purpose. Thus, a common data sharing platform is required where the data producers can
obtain profit from their data and the data consumers can easily find data. To tackle this
a new data marketplace ecosystem is required based on just technical aspects but also the
social and economic aspects. Rather than conventional centralized solutions, our research
focuses on solutions for enabling the data sharing without the stakeholders having to have
full trust in the marketplace owner or provide.
The three main building blocks on this new proposed “Data marketplace ecosystem”
are a community system, open business architecture platform and the relationship between
community system and open business architecture platform. This data marketplace ecosystem
is a decentralized, open and large software system, which is owned, controlled and used by a
community system.
A community system: A community system is a group of people who share a common
interest but still form a heterogeneous system. The community system can be subdivided
into different homogeneous subgroups.
Provide community- The provider community defines some rules and standards for
the ecosystem to function safely. It also helps the ecosystem evolve based on the
requirements of the user community.
18 Z. Abrahamson, “Essential Data,” 2014.
Lucy Handley, “Amazon beats Apple and Google to become the world’s most
valuable brand,” 2019. [Online]. Available:
amazon-beats- apple-and-google-to-become-the-worlds-most-valuable-brand.html [Accessed: 18-
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 123
User community: The users of the data marketplace are the data providers and the
Operator community: The operator community operates the data marketplace ecosys-
tem. The operating community provides the computation and technical infrastructure
for the data marketplace to function. The goal to have such a community avoid single
Open business architecture platform: The open business architecture platform is the
platform i.e. the data marketplace itself which the community systems develop, operate
and use. It describes the technical realization of a whole system for a data marketplace
ecosystem. The main goal here is to provide a completely decentralized data marketplace
ecosystem, which is open and flexible as far as possible but still provides nonfunctional
requirements like safety, security, privacy and dependability.
Relationship between the community system and open business architecture platform:
The community system has various responsibilities which helps defining, development
and the evolvement of the open business architecture platform. These responsibilities are
the relationship between the community system and open business architecture platform.
As the community uses and evolves the ecosystem understanding this relation is very
The proposed concept is community driven and proposes on the one side an initial concept
for the community structure and on the other side an architecture which is open, flexible and
secure and is aligned to this community. Nevertheless, this project is still work in progress
and will be continuously expanded in the near future.
Although such a data ecosystem provides new opportunities for data consumers, providers
and many other stakeholders, there are many challenges to be tackled. We identified various
challenges in the data marketplace ecosystems. Although while writing this, our research is
still work in progress we sketch some solutions for tackling few challenges.
Open and secure infrastructure: In order to provide a fairness and transparency, new
methods for secure but an open infrastructure is required.
Metadata: In order to sell the data on the marketplace the seller needs to provide some
description about the data. E.g. What is the data about, the size of the data etc. This
information about the data is known as meta data. On one hand meta data gives the
buyer the data information and is also used for search which helps the buyers to find
relevant data. But providing such detailed relevant information about the data can be
too much work for the sellers. One possibility for this challenge is Automatic generation
of metadata. But the question still is how to generate the meta data automatically and
protect it from unauthorized access.
Data quality: Data is very different the any other commodity sold on electronic market-
places. When a user buys a physical product online, this product can be returned or
exchanged if he/she is unsatisfied with it. But the same does not applies for datasets.
Once the buyer sees the datasets i.e. buys them, it cannot be returned. Thus, the biggest
challenge we identify for trading data as a commodity is ensuring the data quality.
7.21 Jakob Rehof (TU Dortmund, DE)
When data ecosystems are understood to involve distributed networks of data providers and
data consumers the question arises how to organize data logistics in the sense of a data supply
network. Such issues may reasonably be attributed to the sub-field of data engineering.
124 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
A central data engineering challenge in this context is the automation of data supply
networks, that is, essentially, the task of getting data from A to B in a form understood and
desired by A and B. If two or more parties wanting to share data need to engage in arbitrarily
complicated software engineering projects prior to making data flow among themselves, the
prospect of realizing data ecosystems at large is fundamentally inhibited.
An interesting line of research motivated by these observations concerns the employment of
component-oriented synthesis (much in the sense discussed at the previous Dagstuhl Seminar
14232 on Design and Synthesis from Components) as a means towards automating data supply
chains. In particular, the automatic generation of complex data transformation functions
and rendering functions appears to be a useful goal which should not be too far out of reach
for practical exploitation. Based on some years of experience with research in type-based
component-oriented synthesis in Dortmund, the idea of developing specific frameworks for use
in data supply networks appears natural. The capacity of such frameworks to operate relative
to given (but possibly dynamically changing) repositories of components seems useful here.
For example, data provider A could inject data transformation functions based on domain
specific knowledge of data semantics together with accompanying metadata. Data consumer
B could inject domain specific transformers and rendering functions pertaining to the domain
of use. An interesting fact about such a scenario is that the repository could evolve in a
distributed fashion without any need for a central control other than the logical control
implicit in overarching standards such as may be embodied in metadata formats. Thus a
vision may arise which foresees distributed and (to a large extent) self-organizing networks of
functionalities (the repository) accessible to automatic methods of code generation (synthesis)
composing functions on demand to achieve a stated goal (for example, the transformation
of data from A to B in a desired form). One can go further and imagine such a framework
being combined with query languages and query mechanisms tailored for the network (such
that, e.g., a data transformation function is automatically synthesized as a side-effect of
executing a query against the network).
7.22 Simon Scerri (Fraunhofer IAIS – Sankt Augustin, DE)
Due to the well-understood AI opportunities presenting themselves in the last few years,
there has been a steadily increasing and consistent interest in data sharing methods, solutions
and practices. This has been observed internationally both at an industrial level, as well
as at a political level. Seeing this as an opportunity to boost the data economy in Europe,
the European Commission has reacted strongly by organising many events at attempting to
invest in the convergence of technology and infrastructures that can enable the realisation
and adoption of a pan-European data sharing space that can incorporate existing vertical,
cross-sectoral, personal and industrial data spaces and enable broader participation. The
realisation of an ’open’ data sharing ecosystem that can serve the needs of all kinds of
stakeholders also introduces exciting opportunities for scenarios that are not only restricted
to B2B, but also enable data exchange possibilities for science, government and private
citizens. For this vision to be achieved, the convergence of efforts at both industrial and
socio-political level, through the adoption of standards that respect the existing legal and
regulatory frameworks (e.g. GDPR for personal data), is being encouraged.
In view of this European vision for a data sharing space, the Big Data Value Association,
an industry-driven international not–for-profit organisation with 200 European members
(composed of large, small, and medium-sized industries as well as research and user organiza-
tions) has an activity group following and promoting advances in Data Sharing ecosystems.
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 125
Simon Scerri is one of the lead editors for the position paper published on the topic
. In
its ambition to support the convergence of existing efforts, standards and technology in
this sphere, the group is continuously seeking to collaborate with external experts and data
practitioners. The position paper, for which a second version is planned for early 2020,
includes a survey of the broad (international) technical landscape, and enlists the known
opportunities (for business, science, government and public bodies and citizens), known
challenges (legal compliance, technical, business and organisational, national and regional)
and a list of key recommendations that can pave the way for the targeted convergence
to materialise. The recommendations are addressed to both European policy makers and
industry alike, as the two primary entities identified as having the highest likelihood of
accelerating advances in the area.
The technical concerns and requirements discussed in the Dagstuhl seminar have helped
to both confirm the challenges identified, as well as to bring to the focus additional concerns
and relevant initiatives. In particular, the ’blueprints’ for a high-level generic data sharing
architecture that is open to all will be considered for a similar illustration in the second
version of the BDVA position paper. For further information or updates, please contact
Simon or refer to the BDVA Website Downloads section in the near future. In addition, an
appeal is made for entities with an interest in data sharing practices to refer to a survey
which is due to be published and promoted by the BDVA in January.
7.23 Julian Schütte (Fraunhofer AISEC – München, DE)
Future data ecosystems go beyond centralized cloud services and will require new technical
paradigms and infrastructures to establish trust among participants. Central cloud providers
currently serve not only as Identity Providers (IdP), but also as implicitly trusted data
brokers which are expected to treat sensitive business data confidential, reliably enforce access
constraints, and provide accounting and billing services. However, many upcoming business
cases cannot be served by a single trusted cloud service – either due to legal reasons or due
to technical necessities that require an orchestration of several services and stakeholders.
To overcome the limitations imposed by central cloud services, technical infrastructures
that allow direct data exchange between participants without relying on central cloud services
are needed. This raises various research questions, such as the establishment of trust between
peers, the control of data flowing between services across enterprise boundaries, and ways to
protect data while still being able to process it.
While some of these individual problems have been addressed by the research community
in the past, synergies from recent developments in distributed architectures, in advanced
cryptography, and in upcoming commodity hardware will allow to create a technical found-
ation for data-driven business cases that do not depend on centralized cloud providers. I
would like to make the following contributions:
Discussion of the main security building blocks needed to create trusted decentralized
data ecosystems
Insights into data usage control systems and their technical manifestation
Presentation and discussion of mechanisms and upcoming standards for trust establish-
ment, i.e. the foundations of automatically establishing trust in a remote party at a
technical level
126 19391 – Data Ecosystems: Sovereign Data Exchange among Organizations
Examples on how novel security mechanisms such as cryptographic access control in
decentralized ledgers will foster new business cases in the logistics domain
7.24 Egbert Jan Sol (TNO – Eindhoven, NL)
The evolution of our (mechanical/electronics) industry towards a data driven industry (data
eco-systems) involves many technologies. IoT data collection is one of the many aspects. It
is the basis for coupling data to Digital Twin, AI-algorithm, but also data-driven businesses
as servitization where measured usage of products leads to e.g. billing. IoT data collection
faces several challenges to overcome. There are far too many different (fieldbus type) IoT
data communication standards, limiting data use to local usage. But standardization as
OPC-UA and usage of 5G will lead to an explosion of (large/big) data sets within and across
businesses. Collected IoT data is not copy-right protected as it doesn’t involved creative
labor and needs extra legal and cyber security measurement. For AI usage data must be
cleaned and for sending bills certain IoT data must be treated with DLT (distributed ledger
tech/blockchain tech). And, maybe the biggest problem for business, is a huge lack of skilled,
trained people with the proper digital skills as practically every non-university trained person
above 35 years today didn’t get any training in digital technologies 20 years ago at school
when they were 15 years. With the rapid deployment of Industrie 4.0 mankind faces for the
first time in history the need to life-long retrain every one in digital skills. For me the main
question is what are the new big data/AI etc data technologies that are needed, that will
be developed and how to make them usable, cq how can we educate and train people (and
politicians) at academic, higher level and vocational level to understand them, to create the
proper data-ecosystems and to deploy them. This challenge is not limited to my background
in manufacturing industries, but is similar for food, medical and many other domains.
7.25 Maria-Esther Vidal (TIB – Hannover, DE)
During my academic and profesional work, I have been involved in diverse projects where
the resolution of interoperability and quality issues across large volumes of heterogenoeus
data sources is a pre-condition for effectively devising data integration. Albeit being projects
from different areas, e.g., industry, biomedical, or scholarly communication, data variety
and veracity seemed to be domain-agnostic even thought their resolution required domain
specific knowledge. Motivated by this fact, the definion of generic frameworks able to
identify interoperability issues while allowing for an effective data integration process has
been one of my research topics and one of my interests in attending the seminar on Data
Ecosystems. Currently, I am leading the tasks of integrating clinical data from electronic
health records, gene sequences, and medical images, with open pharmacogenomics data
and scientific publications. Natural language processing and semantics annotations from
controlled vocabularies provide the basis for the fusion of these variety of data sources.
During the seminar, I can contribute describing both challenges and solutions that we have
tackled in the context of projects like iASiS
and BigMedilytics
. I can also discuss, the
approaches that we have considered to address these issues during query processing.
Cinzia Cappiello, Avigdor Gal, Matthias Jarke, and Jakob Rehof 127
1Behzad Abdolmaleki, Karim Baghery, Helger Lipmaa, and Michał Zaj¸ac. A subversion-
resistant SNARK. In Lecture Notes in Computer Science (including subseries Lecture Notes
in Artificial Intelligence and Lecture Notes in Bioinformatics), 2017.
2Ziawasch Abedjan, Lukasz Golab, and Felix Naumann. Data profiling: A tutorial. In
Proceedings of the 2017 ACM International Conference on Management of Data, SIGMOD
Conference 2017, Chicago, IL, USA, May 14-19, 2017, pages 1747–1751, 2017.
3Ron Adner. Match your innovation strategy to your innovation ecosystem. Harvard business
review, 84(4):98, 2006.
4Dimitris Apostolou, Gregoris Mentzas, Bertin Klein, Andreas Abecker, and Wolfgang
Maass. Interorganizational knowledge exchanges. IEEE Intelligent Systems, 23(4):65–74,