The digital transformation of science allows researchers nowadays to expose Research Objects through many di erent publishing channels, so that other interested stakeholders can nd and reuse it. Linked Data is an accepted mean in these meta descriptions to enhance Findability, Accessibility, Interoperability and Reusability (FAIR). But researchers face a large variety of established publishing applications, where they have to select between general-purpose or domain-speci c platforms and user interfaces of varying quality and feature set. In order to improve interoperability aspects, we want to analyze which publishing systems currently exist and to which extent they support Linked Data annotations from the very beginning. We therefore concentrated on research data and conducted a systematic mapping of general-purpose research data management (RDM) systems currently in use, and summarize them in a tabular resource. The obtained results were then evaluated against their current support for semantic, interdisciplinary data annotation and exchange. We show, that a large set of established research data publishing solutions already exists, but that their support for Linked Data is still limited and can be improved.
Scientists are encouraged to publish, share and reuse research data (Open
Research, stimulating Open Data and Open Access). Research data (other terms:
scienti c data, scholarly data) is an essential artifact of scienti c work: It leads to
insights, makes research reproducible and validates ndings. Which also means, it
is an integral part of the results of research. The number of scienti c publications
is growing, and consequently also its underlying data [
The digital transformation of science has o ered new publication channels
for research data, starting with direct data exchange in the past, traditional
self-hosted webserver downloads, up to general-purpose or specialized data
repositories, larger centralized or decentralized platform infrastructures with
corresponding non-pro t or commercial application providers. Examples are CKAN,
DSpace, EUDAT, the Harvard Dataverse, Zenodo and derived systems. These
platforms nowadays provide means to upload and publish research data sets
with a general-purpose, disciplinary or institutional focus. They also have to
satisfy speci c needs of multiple stakeholders in di erent usage scenarios ([
The aim of this study is to analyze the actual support of RDM systems for interdisciplinary data exchange from a user's perspective. We will describe bene ts of semantic annotations for interdisciplinary data exchange and are interested in a systematic approach to get an overview over currently used scienti c data publishing applications. Especially for researchers who are not technically versed, it can be a great e ort to nd an appropriate solution; or they will simply use a tool they are aware of in their knowledge domain or that they were told. This negative experience might also be a reason to keep scientists back from sharing their data or switch to simpler publishing approaches and tools restraining interdisciplinary research data exchange and reuse.
This research is an initial part of the PIROL PhD project [
The rest of the paper re ects these objectives in the following way: Section 2 introduces de nitions for concepts related to interdisciplinary research data management. Section 3 describes the methodology used in this research to identify suggested research data publishing systems and describes the search strategy for reproducibility. The results of our systematic mapping are presented in section 4 and further investigated in section 5. Section 6 contrasts our work from existing studies and section 7 summarizes our results and gives an outlook to future activities. cf. https://www.re3data.org/metrics/software https://ec.europa.eu/info/research-and-innovation/strategy/goalsresearch-and-innovation-policy/open-science/open-science-monitor/
De nitions for research data vary depending on institutions and application
domains. Throughout this paper, we follow the de nition, that research data
\covers in principle any kind of digital artifact that is associated with
scienti c research" [
In our study, we are particularly focused on research data publishing
within RDM life-cycle models, which can be de ned as \making a
research artifact online available, discoverable, peer-reviewable, re-usable
according to given rights, real-time accessible, citable, and interlinked
with its research activity and associated products" [
It was already proven that researchers are willing to publish and share research
data to other scientists [
One possible approach to improve this situation is to extend meta data
descriptions of research data sets with additional information. Linked Data is a
means to map related relevant concepts onto each other [
2. Usage of appropriate ontologies, 3. Description of associated concepts, 4. Reference to related resources
through the use of persistent identi ers, which can realize
FAIR principles for data sharing [
We have the hypothesis, that established research data publishing platforms still use traditional user interfaces that only request basic meta information as text, such as provenance information, missing to also allow sophisticated meta descriptions and linking for interdisciplinary reuse which represents an obstacle for discovery activities by other researchers.
Assuming a scenario, that a scientist of a speci c domain wants to publish data of a research project that is also relevant for other researchers, we want to identify RDM publishing applications currently in use. The resulting summary can help researchers to get an overview on relevant systems and compare their feature set. To facilitate the cross-domain retrieval and reuse of published research data, we further analyze all relevant systems afterwards with a speci c focus on the current support of Linked Data. In this scenario, this particularly means how existing RDM systems help a scientist to describe di erent characteristics of a research dataset in a structured way by reusing existing ontologies and appropriate unique concept identi ers.
A prede ned set of suggested RDM platforms or a processing of data source entries in existing data catalogs was not regarded as appropriate because it could miss other relevant publishing channels being actively in use that were not considered in such a registry in advance.
The following section will provide details on the survey methodology for reproducibility concerns. 3.1
The goal of this comparative study is to get an overview on existing RDM solutions with a special focus on publishing and sharing research data for the interdisciplinary discovery and reuse by other researchers. As so, we will concentrate on general-purpose research data publishing platforms that can be used by any researcher independent of a particular knowledge domain. The study of domain-speci c solutions for describing terminologies would be interesting as well, but is out of scope of this paper. 3.2
We ran a systematic mapping approach on scienti c publications dealing with research data management and data publishing approaches. To select relevant papers, the following criteria were used.
Inclusion Criteria The main criteria for the papers selected were provided by the publishing date and the language the paper was published in: + only papers published between 2008 and 2018 + only papers in English
Furthermore, the papers should focus on at least one of those criteria: + research data management + data publishing or data sharing Exclusion Criteria Papers with a focus on other aspects in the RDM life-cycle such as data collection, data processing and data aggregation were excluded. Also papers about systems which do not exist or are not maintained anymore were excluded.
{ focus other than data publishing or sharing { systems do not exist / are not maintained anymore 3.3
Search Terms Formulating search terms to get relevant results was a challenge, as there is much ambiguity in the vocabulary of the subject's area. Several terms are used as quasi-synonyms, e.g., data platform vs. data repository vs. data management system or research data vs. scienti c data vs. scholarly data vs. datasets in general. Nevertheless, the terms had to be limited due to the unmanageable amount of results in a test run. The search was conducted within the title, the abstract and the keywords (if exist), using the following terms: { platform, repository, data management system { scienti c data, research data, datasets
{ publishing { sharing and publishing, the
engines, platforms and digital libraries were used: Google Scholar, ISI Web of Science, ACM Digital Library, IEEE Xplore Digital Library, Springer Link, Mendeley, Science Direct , Emerald Insight, SAGE Journals. Due to the huge amount of results with descending relevance in Google Scholar (21,100) and Springer Link (21,031), only the rst 1,000 results were screened by the title. 4
Conducting the search lead to 865 results after removing the duplicates. The outcome was then reassessed with reference to the criteria, which resulted in a sum of 487 papers. We reviewed all relevant publications and derived a basic systematization, before we further analyze and discuss all solutions with a focus on semantic data management. 4.1
The publications were classi ed in two main categories: rst, general papers including publishing and sharing research data aspects, and second, systems of di erent type to publish and share research data.
The group of general publications on publishing and sharing research allowed us to get an impression on the researcher's behavior and of systems signi cant in the publishing process. The second group of literature contained papers on one particular system. This included software to manage data repositories but also the research data environment of academic institutions. Even though only one system was addressed, quite often related systems were named. All mentioned unique data publishing approaches were noted.
The resulting set of available applications is depicted in table 1. The table contains software mentioned for data publishing purposes independent of the particular application focus. It is notable, that several systems are modi cations or forks of another basic platform. Examples are Dryad which is a platform for journal data submissions built upon DSpace, Zenodo or the EUDAT B2 services built upon Invenio, or Linkitup built upon Figshare. A detailed explanation of each identi ed data publishing solution is omitted as we refer to the corresponding website and other surveys for details.
Interestingly, there were also data publishing solutions from related studies that did not show up in our study. Examples are ContentDM, dLibra, Digital Commons, ESciDoc, Mendeley Data, Nesstar, RADAR, ResearchGate or Springer Nature. They either target a di erent user segment, are placed under di erent search term strategy or are data provisioning channels not mentioned in the actual paper.
We manually removed systems that were knowledge-domain speci c such as AmeriFlux, Arctic Data Centre (ADC), Brain-CODE, COINS, Curatr, ESSD, FID Sociology, GeoNode, GWATCH, IS-EPOS, LearnSphere, National Sleep Research Resource (NSRR), NDA/NDAR, OCHEM, OMEGA, PANGAEA, PCORnet, Phenomics Ontology Driven Data (PODD), reefgenomics.org, SchizConnect, SICAS Medical Image Repository, Sloan Digital Sky Survey, TypeCraft, Scratchpads, VPHShare, Waveform ECG, WorldMap, xiSPEC, XNAT or XTENS ; or that were institutional-speci c instances of an independent project such as DepositOnce as an application of DSpace at the Technical University of Berlin, INFN OAR as a clone of Zenodo in Italy, and TIND or WEKO 3 as an Invenio instance for research data of multiple universities in the US and Japan.
In this section, we discuss the obtained results of the systematic mapping and further investigate all identi ed solution with respect to their current support for Linked Data operations to improve interdisciplinary retrieval and reuse. 5.1
The landscape of solutions to support publishing and sharing research data is hallmarked by a huge diversion, especially through the variety of domain-speci c specialized platforms. Researchers should become aware of the systems that exist and research should focus on lowering barriers between these systems to eliminate data silos and increase interoperability. Gaining a fast impression of a system's functionality was often di cult due to a lack of user-focused self-description and list of features. A better communication would already open them for a bigger community.
It is remarkable, that research data publishing activities often take place via GitHub (software links explicitly excluded). The platform is referenced more often than the Dryad journal platform or the underlying DSpace system which implies a notable current relevance for the research community.
Furthermore, the study revealed publishing channels that are traditional and reasonable, but which were not expected in advance. Examples are the data publication via a website Content Management Systems (CMS) such as Drupal, or the usage of commercial Cloud providers such as Dropbox or Google Drive and similar solutions. As we were looking for particular research data management software, we did not emphasize these research-external solutions, but use it as an indicator that users might stick to accustomed, easy-to-use solutions for le sharing, even if they might be commercially driven or privacy-insensitive. 5.2
The applied systematic mapping methodology had some implications to be considered. Even though the search terms were limited, the amount of obtained search results was huge as many publications nowadays include a data publishing aspect. The applied search result pruning implies, that some systems might have been left out. The exclusion of processing and analyzing might have excluded some very interesting systems with a Linked Data focus. In contrast, the general terms platform or repository often resulted in small and very speci c systems which had no huge value for the general identi cation of research data publishing systems. Also, the large amount of domain-speci c systems and platforms did not allow to validate each solution and check its feature set. We therefore concentrated on domainindependent research data management platforms and analyzed only approaches that showed up at least 10 times or were base systems or higher ranked solutions.
Identifying the main representatives by occurrence can be an indicator to show their relevance and establishment, but not consequently their eligibility and quality. This applies especially for newer and hyped systems. Also, multiple RDM systems did not exist anymore and therefore not showed up in the mapping summary. 5.3
To improve scienti c discoverability, exchange and reuse among multiple research
disciplines, we further investigated the identi ed research data publishing
solutions for the implementation of sustainable interdisciplinary data management
capabilities. In order to compare these RDM systems on assessable
characteristics, we derived 15 criteria C from the FAIR principles [
C1 Is a particular research data set in a current version accessible via a unique PID? C2 Is the research data information through that platform indexed in data catalogs, registries and search engines? C3 Is a search interface available with lter possibilities for structured Linked Data? C4 Can new research data be stored or referenced in an easy way? C5 Is the user input interface Linked Data - aware and easy to use by hiding technical terms and identi ers? C6 Can the research data and/or meta data be accessed directly via http(s)? C7 Do authentication and authorization settings for public/private/restricted access exist?
C8 Is the meta data description available in an RDF serialization? C9 Can particular established ontologies be used to describe the research data set in a general way such as schema.org/Dataset, DataCite or DCAT /DublinCore? C10 Can domain-speci c vocabularies be used to further describe the research data? C11 Can each concept related to the research data set be described with a corresponding resource
URI?
C12 Can a data license be speci ed in a Linked Data fashion? C13 Can the data provenance be speci ed and updated in a structured way? C14 Are data sets set into relationship based on Linked Data and criteria such as the topic, community, used methods or similar? C15 Is the provided data validated or do compliance checks exist?
In order to assess every criterion, a demo instance of each approach was examined together with the provided application self-documentation. We therefore used the basic version without any enabled plugins or sophisticated con guration.
A scoring system was applied and the following symbols were used: + was assigned, if the criterion was entirely ful lled o was assigned, if the criterion was partially ful lled - was assigned, if the criterion was not ful lled % was assigned, if the criterion was not applicable ? was assigned, if it was not possible to assess the mentioned criterion () was assigned, if the feature is limited in the native version but might be there with plugins Table 2 lists the results of this assessment. We thereby categorize the previously identi ed RDM solutions in three types: RDM systems that serve as a basic data management platform, RDM applications as a further development of these platforms, and other tools for data management; as a direct comparison between solutions of di erent application focus appeared inappropriate.
We faced a lack of Linked Data support in the identi ed RDM systems. They
normally provide direct web access to the research data itself via a uniform,
persistent resource identi er (C1, C6), manage a research upload (C4) and
request in an input form additional meta information that is mapped to a
DCAT, Dublin Core or schema.org vocabulary. However, most of the meta
information is still based on literal string input (C5). This means, that the
strength of Linked Data to set other concepts into a comprehensive, distinct
relationship is not used at all. This situation could be improved for user input
interfaces in the frontend by providing recommended and more sophisticated
input operations [
The generated meta information can only be reviewed on every RDM platform and the provided export format is only in a few cases available in an RDF serialization (C8). Provided metadata was mostly limited to basic information and provenance data (C9). Other relevant, domain-speci c vocabularies are seldom taken into consideration or o ered in the user interface (C10). If such a structured research data description is possible, it was limited to a rudimentary provision of property URIs and object values to identify particular topics.
Managing versions and provenance aspects was a key feature, where the majority of the tested RDM solutions succeeded (C13). However, this provenance information was seldom part of the meta description export (e.g., by relying on the PROV-O ontology) which could be of high relevance for reusing research data and existing meta information. Automatic data quality assessment checks were not run by any of the tested systems and only concentrated on a valid meta data input (C15). Instead, platform providers include a manual review process with a con rmation step before publishing a submitted resource.
Outstanding solutions with Linked Data awareness that should be highlighted and considered are currently the Harvard Dataverse, the EUDAT B2 services, Zenodo and DSpace with TripleStore and RDF extensions and/or a Fedora sublayer. Toolwise, Dendro has to be pointed out as a tool to specify potential relevant meta data.
Research data management systems were already compared in several studies.
Kokorceny and Bodnarova [
Poole [
However, all of these surveys assumed a pre-de ned basic set of relevant systems to evaluate. We reassessed the establishment of RDM solutions in 2019 by focusing on an analysis of existing publications and a systematic mapping. Furthermore, we particularly highlight the role and support of Linked Data in these data management systems. 7
In this paper, we investigated current research data management platforms and their capabilities to request and expose Linked Data descriptions of research data sets for interdisciplinary discovery and reuse. We therefore conducted a comparative study where we identi ed 18 web-based general-purpose RDM platforms that are used in 2018 to publish research data.
All identi ed web applications were then evaluated with respect to their current support for dealing with Linked Data to follow FAIR principles for data sharing. We came to the conclusion, that the support for Linked Data di ers among all examined systems and that there is room for improvement, especially from a UI input point of view. Nevertheless, a basic support is already given in systems such as the Dataverse, EUDAT B2 Services, Zenodo or DSpace. In particular, this is limited to discovery meta information with links to author's ORCIDs and basic DublinCore properties in most of the cases. Although dedicated ontologies with unique concept identi ers already exist in multiple knowledge domains that can be used to describe relevant aspects of a data set, they are hardly used due to a lack of appropriate user interfaces and terminology services.
A closer look, especially on the huge variety in knowledge domain-speci c systems, could be a bene t for interdisciplinary research and for the exchange of experiences and best practices