In the age of information technology and Linked Data, the importance of creating transparent infrastructures for management and exchange of data have emerged to play a focal role for the development of new open research. The FAIR guiding principles have been introduced to advise in the improvement of Findable, Accessible, Interoperable and Reusable technical resources, but the literature is still lacking of practical guidelines for the management and digitization of sensitive and restricted data. The purpose of this research is to introduce new insights aimed at the exploration and discovery of restricted data-sets. The methodology sets out by creating a model for FAIR metadata architecture, which will later be used for automatic ingestion of external restricted data-set les, as well metadata enrichment. Quantitatively and qualitatively evaluations will be performed to assess the work ow, which will later be implemented in the data-set's search engine. Initial results of the metadata ingestion and enrichment are presented as the RML mapper rules and the OPL ontology, respectively.
In the era of digitalisation and Open Science [
Many archives in the biomedical domain have already made their data open
and ndable, such as GenBank [
Due to the sensitive nature of a large number of available data-sets, a growing
number of initiatives are in facts developing research tools based on the top-level
information, the metadata. Nevertheless, formal guidelines for the management
and digitization of restricted data and metadata creation is still lacking.
Moreover, a broader perspective regarding the use of detailed metadata les as a
in uential factor in di erential privacy and privacy budgets supported queries
[
The term \metadata" is generally understood to mean \data about data", and
it usually describes resources' embedded information such as authors, dates,
versions and technical details [
It is clear that such information are truly valuable resources to allow researchers for a more targeted data-set search and exploration, and there is an evident need for such knowledge to be available to the end user. 2
A more detailed account of relevant researches are described in the next section. Firstly, we will discuss the transformation from raw data into Linked Data with the RDF Mapping Language (RML). Afterwards, we will present the Open Digital Rights Language (ODRL) and the CESSDA Metadata Model (CMM), which represent the core frameworks for the novel OPL ontology and the metadata architecture model respectively.
RML. An important step in this research is the creation of Linked Data from
unstructured metadata les. For this step, the RDF Mapping Language (RML)
[
ODRL. The Open Digital Rights Language [
CMM. The Consortium of European Social Science Data Archives (CESSDA)
has made extensive e orts in the introduction of various guidelines for data
management and maintenance, by highlighting the importance of structured
metadata in the social science domain. One of the core activity of the CMO (CESSDA
Metadata O ce) has been supporting the data-origination process by
establishing CMM [
The main question of this study is: How can restricted data be FAIRi ed?. This study will focus on social science data, and more speci cally on tabular data, but its approach could be extended to any eld of research. The assessment of this speci c objective consists in the evaluation of the following sub-questions:
RQ1.How can we represent tabular data that comes with access restrictions in a FAIR manner? The rst step in applying the FAIR principle on restricted research data is the creation of a model or system architecture that is both open and transparent, as well as consistent with the requirements essential to the research community. Due to the constraints imposed by the context of this resource, we will focus on a metadata model that expresses information about the underlying data, therefore avoiding con dential material being exposed.
RQ2.How can existing scienti c data-sets with restricted access be FAIRi ed in a reproducible and transparent fashion? The second step consists in the practical evaluation of the approach. In order to achieve this, we aim to map and enrich the metadata of available data-sets, from a raw format into their Linked Data counterparts.
RQ3.How can researchers dealing with access-restricted data be supported by semantic tools to create data-sets that are FAIR from the start? We want to support the research community and data providers in the generation of metadata for restricted tabular data, by creating a transparent tool based on the model proposed. The tool will support the use of semantic web technologies in to order to facilitate logic-based annotations, as well decrease disambiguation in entity-recognition processes.
RQ4.How can FAIR metadata for access-restricted data improve the e ectiveness and usability of a scienti c data-set search engine? The end goal of this study is to provide a data-set semantic search portal, where users are able to nd resources by querying the metadata database. Therefore, our nal question for this research, involves the comparison between the performance of the data-set search portal created and other available search environments. Moreover, we will evaluate whether certain implementations in the semantic search of the portal are more bene cial and powerful than others, in order to make suggestions for future work and progress. 3.1
The contributions that this research is set to implement are: { The formulation of a comprehensive metadata model suitable for restricted scienti c tabular data, with the input of eld experts in the social science domain. { A method for generating enriched and FAIR metadata by applying the model to existing restricted data-sets. { The creation of an automated yet transparent and FAIR tool for the generation of Linked Metadata from the upload of data-sets. { The contribution to a data-set search portal, by implementing semantic search features to optimise the ndability of resources. 4
The overall approach of this research is based on the principles of: metadata architecture and enrichment, con dentiality-aware data processing, semantic rights representation and access negotiation. Having de ned the research questions, we will now address the methodology in more details:
RQ1. A systematic review will be performed to assess, summarise and
determine the relevant literature in regards to the management of con dential
data-sets, as well as common practices for the exploration and use of restricted
data. The outcome of the review will be translated into the metadata model'
requirements, and compared to already available resources such as the CMM
model, the RDF Discovery Vocabulary (disco) [
RQ2. Raw metadata can be obtained by di erent data providers (e.g. CBS
and DANS), and can be transformed into Linked Metadata following the model
mentioned, using known technologies such as RML. Nevertheless, during this
process original information are kept intact, and no extra knowledge is derived.
In order to enrich the metadata we can utilise available vocabularies linkable
to the original data, such as the DBpedia Ontology [
RQ3. A novel tool is needed to automatize the generation of metadata, therefore supporting restricted-data owners in the creation and release of their resources into digital libraries and metadata portals. Semantic tools, such as entity-recognition and recommendation services will be put in place to map the data-set terms to known thesaurus, such as ELSST, and users will also have the option to customise the terms by adding or suggesting a more appropriate mapping. We aim to create a secure and transparent environment where data owners are in full control over their data, and where the tool work ow is open and accessible to the users, in order to allow them to check every stage of the transformation and decision-making processes.
RQ4. Firstly, semantic search results among di erent portals and the proposed data-set search engine will be compared, in order to better understand its performance in nding information. Moreover, we aim to analyse the ndability of resources, by using two versions of the novel designed portal: 1) this version will be considered to be the \control" group, and it will consist of the portal with all the search functionalities running, 2) this version, instead, will be considered as the \knock-out" group, where certain search functionalities are disabled. The experiment aims to understand which parameters are more powerful and, therefore, to suggested further areas of study. Finally, we will evaluate whether the FAIR implementation in the metadata model allow users to nd restricted data-sets, as well as clearly stating the steps and requirements needed to be taken in order to have access to the data from the data-owner. A possible addition to this methodology, is also to investigate privacy budgets supported queries, by tracking users' history and therefore calculating the level of security of the portal.
5
Having de ned the research questions and the methodology of this research, in the following section we will discuss how to evaluate the results.
RQ1. Following the systematic review, metadata requirements found in the literature will be assessed and implemented in the metadata model, and major importance will be given to literature exploring the Reusability and Findability of restricted data. The focus group will be eventually be questioned regarding the nal version of the model.
RQ2. A collection of open data-sets in raw format will be used to manually generate their correspondent metadata les. A case study will be performed to quantitatively evaluate the results by counting the number of terms that have been correctly mapped to the model, as well as counting the triples before and after the metadata enrichment. Once the metadata creation is optimised, restricted data-sets from providers such as CBS will be requested and investigated.
RQ3. The tool will be tested by feeding it a collection of data-sets for the automatic mapping to the metadata model, with the aim to achieve the same performance as in the manual mapping mentioned above. Moreover, a focus group consisting of open- and restricted-data owners, as well as researchers, will be asked to perform the task extensively, and their feedback will be assessed.
RQ4. To evaluate the last research questions, we will organise a comparative user study to examine the features implemented in the last iteration of the proposed portal, compared to other available resources. Moreover, we will check whether all requirements found during the systematic review, and further feedback that may arise during this study have been integrated. Furthermore, we will qualitatively assess the strength of semantic search features, by checking the resources returned by the \control" version and the \knock-out" version. Privacy budget queries could also be evaluated by collecting analytical information about the search history of users, and specify a clear privacy cost limit. Finally, the aim of the portal is to include metadata from restricted data-sets, and make them Findable for the research community. 6
To create an initial sample for experimentation, RML was used to create a mapping of XML les from the DANSeasy archive, into RDF Linked Data format. RML default functions were used to enrich the metadata, by mapping XML access terms to the OPL ontology. Therefore, we have initials answers to research questions 1 and 2: { RQ1: during the RDF transformation with RML, the CMM model has been used as a template. We have seen that this model is exible and expressive enough for initial experiments, but further discussion with the research group are needed to evaluate potential implementations. { RQ2: metadata enrichment has been performed by mapping XML terms to OPL, and initial feedback shows that a number of additional information about license agreements and data usage have been added. This step is important in the FAIRi cation of the model, as the aim is to provide end users with transparent and e ective information about the data. The code for the XML mapping can be found at 3, and for OPL ontology at 4. 7
The present study was designed to underline a model for FAIRi cation of restricted research data. Although this research has only started at the beginning of 2021, early ndings have signi cant implications in the rst and second research questions, regarding the syntactic and semantic features of the model. In fact, we have shown current metadata models (e.g. CMM) and resources (e.g. ODRL) can be expanded to optimise ndability of data. Moreover, early implementations of the OPL ontology have shown how license agreements' details can express important features necessary to the research community. Furthermore, we have acknowledged the usefulness of available tools such as RML during the transformation from raw to structured metadata, and we have also shown how such tools are exible enough to allow the creation of transparent custom functions to better extract and map information.
Considerable work is needed to gain access to restricted data-sets from providers such as CBS, and a great focus is indeed necessary to guarantee the con dentiality and e ectiveness of the model. Notwithstanding these limitation, the aim of this study is to strengthens the importance of metadata and application of the FAIR principles as solutions to restricted data-sets search, and although it focuses on the eld of social science, the ndings may well have signi cant implications in other scienti c communities.
Acknowledgements. This project is funded by the Netherlands Organisation of Scienti c Research (NWO), ODISSEI Roadmap project: 184.035.014. Furthermore, I would like to thank Jacco van Ossenbruggen, Tobias Kuhn and Ronald Siebes for the help and support needed for the success of this research and its future endeavors.