Effective data governance and management are necessary but challenging prerequisites for creating value from data assets. Findability, accessibility, interoperability, and reusability are guiding principles for data owners in managing and archiving datasets, known as the FAIR Principles. Dataspaces provide an infrastructure for heterogeneous, multi-source data integration and cross-organizational data sharing that would benefit from FAIR compliance. In this paper, we propose semantics as an approach to ensure data FAIRness, enabling machine-aided discovery and reuse of data in different formats and structures. We conduct a systematic literature review to translate the overarching principles into ten concrete methods that can be implemented using semantic technologies. In addition, we analyze three mature dataspace initiatives for their adherence to the FAIR Principles and describe their specific implementation. In summary, we argue that semantics provide a common and infrastructure-independent foundation for data management in emerging dataspaces.
Data is a valuable strategic resource for competitiveness, driving innovation and the digital transformation of organizations. Business value is created by using and reusing data assets [1]. Consequently, organizations across all industries are adapting their strategies to incorporate data into their processes and take advantage of opportunities for optimization and automation [2]. Relevant data may originate from various sources belonging to different actors in the supply chain or from other market participants, which could also include competitors. Therefore, data-driven value is often created by combining datasets of multiple actors [3]. By facilitating cross-organizational data sharing, dataspaces are becoming increasingly important for value cocreation from distributed data [4,5]. Their potential is also being recognized at a political level, and governmental funding for dataspace initiatives such as Gaia-X encourages the development of secure infrastructures for data sharing [6].
Handling the increasing volume of unstructured data necessitates proper data governance and management practices. Typically, decisions on their concrete implementation are left to the data owner and remain undefined [7]. This heterogeneity poses a major challenge for dataspaces. Therefore, various dataspace initiatives agree on the four principles of findability, accessibility, interoperability, and reuse (FAIR) as common data management practices to enable smooth integration of multi-source data [8]. However, the FAIR Principles are described at a general level and allow for various implementation options, complicating standardized data management. Semantics enrich data with context and thus enable automated search and processing. Using common, established semantic vocabularies can support FAIRness -i.e., compliance with the FAIR Data Principles -in dataspaces [8,9]. Previous literature deals with individual aspects of FAIR that need to be synthesized to provide a guideline for data management in dataspaces. Therefore, we aim to answer the following research question: How can semantics contribute to FAIR compliance in dataspaces?
The remainder of the paper is structured as follows. First, we detail the concept of dataspaces, semantics, and the FAIR Principles. Second, we describe our methodological approach. Third, we present the semantic approach we identified for each FAIR Principle. Fourth, we assess three mature dataspace initiatives for their FAIR compliance. Finally, we conclude with a discussion of our findings.
Dataspaces. Database management systems are limited to structured data that correspond to a predefined schema. Consequently, these traditional systems are unsuitable for integrating and processing the increasing volume of heterogeneous data coming from multiple data sources [9,10]. Franklin et al. [11] describe dataspaces as an abstraction for managing data regardless of its format or structure. Unlike database management systems, dataspaces do not force the complete integration of data. Instead, they adopt an incremental integration approach and gradually improve data accessibility and interoperability by leveraging Semantic Web technologies for metadata management [11,12]. Recently, dataspaces have been attributed immense potential for value co-creation from distributed data, providing a decentralized infrastructure for crossorganizational data sharing while maintaining data sovereignty [4,13]. Large-scale, governmentfunded initiatives such as Gaia-X1 aim to establish a standardized platform that facilitates data sharing and minimizes the effort required to integrate multi-source datasets. Repetitive and low-level administrative tasks such as search functionality, naming conventions, data lineage, and access management should be reduced and streamlined [9,11]. So far, the dataspaces initiatives are still in their infancy and standards have yet to be defined. FAIR Data Principles. Data governance and management are essential for the reusability of datasets. In many cases, individual procedures are applied that remain undefined and opaque, thus complicating or preventing data reuse by third parties. Wilkinson et al. [7] define four foundational principles, known as the FAIR data principles, to standardize inconsistent data management and archiving practices in the scientific environment. Within the FAIRification process, digital resources become manageable. The underlying four principles are interconnected but can be implemented independently of each other. Besides the actual data, the process also includes the algorithms, tools, and workflows relevant to data collection. Overall, compliance with the FAIR Principles reduces human intervention by enabling machines to automatically discover, process, and integrate data [7,14]. Efficient data management in line with the FAIR Principles is considered particularly important for dataspaces to facilitate data sharing [15].
A common language for describing data assets is crucial for managing and sharing data in dataspaces [10]. Semantics are predefined vocabularies to describe relationships between data entities in a machine-readable format, supporting data integration and enabling automated data discovery [16]. The variety of terminologies poses a critical challenge for communication and uniform understanding between different parties [10]. Seamless data integration from multiple sources requires semantic interoperability, i.e., a shared understanding of the vocabulary. Ontologies aim to reduce semantic heterogeneity through standardized and generally accepted vocabularies [17].
This paper investigates and answers the research question through a systematic literature review. The process is based on the structures proposed by vom Brocke et. al [18] and Webster & Watson [19], which describe the path from general to specific results. Following vom Brocke et. al [18], we begin our research by conceptualizing the topic and delineating the search fields: 'dataspaces', 'FAIR Data Principles', and 'semantics'. The purpose of this study is to examine the correlation between semantics and FAIR dataspaces. To achieve this, we consider each of the FAIR Data Principles separately. Thus, we use the search string: "dataspace*" OR "data Space*" AND ("FAIR" OR "find*" OR "access*" OR "interoper*" OR "reuse*") AND "semantic*". In the second step, we search the following databases for relevant literature: ACM Digital Library, AIS eLibrary, IEEE Xplore, and Scopus. For ACM and Scopus, we limit the search to the title, keywords, and abstract. The search was conducted in February 2024. ACM yielded n=30 articles, AIS n=58, IEEE Xplore n=225, and Scopus n=163 articles. We evaluate the identified articles in an initial review by scanning the title, abstract, and keywords for relevance. Articles that do not address any of our defined search fields are excluded. Duplicates and non-peer-reviewed articles are also removed. After this first phase, we are left with n=78 articles. We then perform an in-depth screening of the abstract, methodology, results, discussion, and conclusion, removing those that do not address synergies between semantics and FAIR dataspaces. After this screening, n=26 articles remain. In addition, we perform a forward and backward search as proposed in [19], resulting in n=21 additional articles. These articles are analyzed in the same cycle as those retrieved directly from the search string, resulting in a total of n=37 articles.
Findability. To ensure findability in dataspaces, three requirements can be met by semantics: self-descriptions, metadata, and catalogs. Digital resources require globally unique and persistent identifiers (PIDs), such as Uniform Resource Identifiers, specified in RFC 3986 2 . These identifiers are widely used in current initiatives, such as Gaia-X, allowing resources to be addressed, modified, and shared separately, even across multiple dataspaces [8]. In addition to PIDs, the self-description of resources also includes information about the entities' attributes and relationships, often using subject-predicate-object triples to comply with metadata [20,21]. Therefore, the Resource Description Framework (RDF) data model is commonly used to model these data structures. Using RDF vocabularies, such as the Data Catalog Vocabulary (DCAT) 3 , dataspace participants can exchange standardized information, including asset name, type, and lineage. This results in data structures that are both human and machine actionable. Catalogs and search tools, such as the RDF query language SPARQL, allow users to access shared information and browse descriptions to find valuable data for their specific use case [22].
Accessibility. Semantics are crucial for accessibility in FAIR-compliant dataspaces, particularly regarding authentication, authorization, and pipeline. Like assets, dataspace participants receive a unique and descriptive identity, managed and verified by external identity providers. Trust between identities is ensured through cryptographic verification, which allows only authenticated actors to engage within the dataspace [23,24]. Due to the sensitivity of data, global sharing is not desired. Authorization for access and usage rights is required. Thus, participants manage their offerings, data or services, via policy frameworks to determine the use and scope of data. Standards such as the Open Digital Rights Language (ODRL) facilitate the creation, communication, and execution of contracts in the data space [25,26]. Leveraging semantics enables machines to understand interfaces and processes, and thus allows for automation [27,28].
Interoperability. Dataspaces promote the exchange of data between parties. To ensure a seamless process, dataspaces must be interoperable. This requires participants to exchange data to be able to comprehend and integrate them. Semantics can aid in standardization and integration. Usually, dataspaces offer a vocabulary hub, providing an overview of commonly established and standardized terms and vocabularies as examples and best practices for describing data, services, and contracts [37,46]. The hub facilitates the exchange of documentation among relevant parties, promoting a shared understanding [29]. Therefore, it enables the integration of data from different sources, such as multiple datasets in a structured format (e.g., RDF), which can be aggregated and queried together. This allows to reveal previously unknown connections between data [50,51].
Reusability. All of the previously mentioned principles contribute to the concept of reusability. For example, if an entity is easily findable, it is more likely to be reused. Similarly, easily comprehensible entities by following a common standard increase the likelihood of reuse. Additionally, semantics can improve the reliability and enrichment of data. Reliability assessment and assurance can be achieved by using shapes to verify the format of incoming data and automating standardization, as mentioned under interoperability. RDF validation is often performed against shapes defined in the Shapes Constraint Language (SHACL) or Shape Expressions (ShEx), which define the requirements that an RDF graph must meet [53,54]. Enrichment can be applied throughout the data lifecycle, mapping raw data into a structured format and augmenting it with additional information from external sources [45].
Several endeavors aim to define principles and guidelines for the creation and development of dataspaces and to transfer concepts from theory into practice. Today, several well-known initiatives in various application areas have reached distinct stages of development and maturity. Initially, such initiatives define the requirements, principles and specifications to be considered when implementing dataspaces. Subsequently, these specifications can lead to implementations or elements that facilitate the establishment of dataspaces in which different actors or participants can exchange data. These implementations can then be sector-specific, such as Catena-X4 in the automotive sector, in Industry 4.05 with a focus on manufacturing, Prometheus-X6 in the education and skill sector, or with more general developments such as FAIR Data Spaces7 at the interface between research and various application domains.
It is also worth mentioning that additional projects looking to facilitate the implementation of dataspaces exist, such as the Eclipse Dataspace Components (EDC) 8 , whose adoption in several projects is documented. The EDC is a framework for data sharing in a cross-organizational and sovereign manner supporting specifications from IDS and Gaia-X. One of its main components is the EDC Connector, a well-defined interface exposing the dataspace participants' back-office infrastructure, aiming to bring together their otherwise incompatible and without access/usage control infrastructure. The connector provides functionalities such as discovering, connecting, automated contract negotiation, policy enforcement, and auditing processes. The Minimum Viable Dataspace9 is a sample implementation of the EDC, leveraging it and showing its capabilities. Currently, these two do not support semantics, but one could extend the connector to include such functionality. Other implementations of connectors exist, such as those listed in [55]. The Sovity Connector10 , based on the EDC and the Dataspace Connector, extends the functionality of the EDC. For example, it allows the EDC to communicate with the catalog called IDS Metadata Broker 11 .
In this paper, we solely focus on assessing the leading and most frequently mentioned initiatives in research. The most commonly noted initiatives in research include Gaia-X, International Data Spaces (IDS), and the European Open Science Cloud (EOSC). These initiatives define the architectural frameworks for dataspaces [56]. Our assessment of the initiatives consists of reviewing the official documentation of the specifications and the repositories with available code. In this way, we want to find out whether the written specifications reflect concrete implementations and how these align with the dataspace requirements introduced in Table 1. The results of our assessment are shown in Table 2. There, we summarize our findings as follows: the symbol ✓ indicates that the requirement is fully implemented, (✓) shows that the requirement is only partially supported, i.e., it is only present as a specification but not yet implemented. However, a * indicates that the requirement is not even part of the specification.
The evaluated FAIRness of current dataspace initiatives
Gaia-X. Gaia-X aims to establish an ecosystem, whereby data is shared and made available in a trustworthy environment. To achieve this goal, Gaia-X defines several federation services, where a federation implements and federates a dataspace. These services are grouped into four sets, namely Identity & Trust, Federated Catalogue, Sovereign Data Exchange, and Compliance12 . One major reference implementation of federation services is provided by the XFSC (Cross Federation Service Components) repositories under the Eclipse Foundation 13 . The Identity & Trust set of services contains an Authentication/Authorization service, which implements these requirements as defined in Table 1. The Federated Catalogue set consists of the Federated Catalogue as well as tool support for self-descriptions. The implementation of the XFSC Federated Catalogue enables the management of self-descriptions and also allows for the validation against SHACL shapes as a measure towards reliability. The topic of Metadata and Standardization are addressed in the Gaia-X Trust Framework 14 . In this framework, a Gaia-X ontology is specified, which has to be used to describe all participants and services of a Gaia-X dataspace. Additionally, constraints are specified, which are modelled using SHACL. All ontology and SHACL graphs can be accessed using the Gaia-X Registry 15 . The Gaia-X specifications also specify the re-use of certain metadata vocabularies such as DCAT, for example, for the data exchange services 16 . The topic of integration is partly specified by the notion of service compositions 17 , which allow the aggregation of several services. This way, a service can re-use data that emerged from the application of another service to apply further processing steps. the remaining requirements defined in Table 1, namely pipeline and enrichment, are currently not addressed by Gaia-X specifications or implementations.
The latest developments of the IDS Dataspace Protocol 18 reflect the specifications regarding the Dataspace Information Model, which covers the definitions of the main concepts to be considered in IDS-based dataspaces. The implementation of connectors in an IDS dataspace context must respond with a JSON-LD data object compliant with JSON Schemas and SHACL shapes. Moreover, participants describe themselves and their resources and infrastructure. These self-descriptions can be registered, published, queried, and maintained by the IDS Metadata Broker. The self-descriptions are metadata, and the use of existing semantic web standards is favored. In this sense, the IDS Information Model [23] is modeled as an RDF/OWL ontology and reuses concepts of the DCAT, ODRL, Time, DQV, and other vocabularies 19 . Moreover, SHACL shapes are available for testing 20 . There is also an implementation of the Dataspace Connector 21 , which uses the IDS Messaging Services for the functionalities and message handling, as specified in the IDS Reference Architecture Model 4.0 22 and integrates the IDS Information Model. Moreover, the Catalog Protocol specifies how a data consumer requests a catalog from a catalog service. Such a catalog is DCAT and ODRL compliant. In the latest version of IDS RAM 4.0, the Identity Authority role includes specifications about the authorization functionalities and the clearing house 23 , which serves as an intermediary to provide clearing and settling services for the data exchange transactions in the IDS. Such a component is similar to the Data Exchange Logging Service from Gaia-X 24 . Additionally, the Dynamic Attribute Provisioning Service 25 is part of the Identity Provider to verify the attributes of the participants and connectors in the dataspace.
The EOSC initiative addresses the FAIR Principles, with interoperability as a core concept, and aims to create a shared data space for research, science and innovation data management while ensuring the protection of data through EU laws [57]. The specified requirements regarding semantic interoperability are the following: there should be a definition of the concepts, their metadata and data schemas, and they should be publicly available; semantic artefacts should be FAIR, available preferably using open licenses, and have associated documentation, and support maintenance; a metadata model, based on existing standards, should be available to allow discovery over existing federated research data and metadata; there should be building blocks and protocols to facilitate the federation and harvesting of semantic artefacts catalogs.
The technical layer defines a common security and privacy framework covering authorization and authentication functionalities. The FAIRCORE4EOSC 26 project focuses on the development of core components 27 Summary. Table 2 shows that two of the initiatives, Gaia-X and IDS, satisfy most of the requirements for semantics, while EOSC has specifications for such requirements, but they are not all implemented. It also shows that these initiatives mostly lack specifications for the more specific functionalities such as pipeline, integration, and enrichment requirements for dataspaces, except EOSC which offers more advanced features covering research discovery through their metadata. Such findings indicate that the initiatives are currently focused on defining the main elements of dataspaces. These efforts cover aspects like authentication and authorization, offering a catalog of available resources for transfer, ensuring self-descriptions and metadata to identify these resources, promoting standardization by using existing (W3C) standards, and providing SHACL shapes for validation and reliability. Another important remark, based on our research of the repositories of the initiatives and connectors, is that some IDS components, such as the IDS Connector are not currently maintained in their original repositories. However, Sovity is supporting the maintenance of the IDS Connector, and extending some of its functionalities. Regarding Gaia-X, some specifications and implementations have not been updated since their initial release, e.g., the implementations of the XFSC components implement the specifications given by the 21.03 version of the Trust Framework, which has since then been replaced by the newer 22.10 version 28 .
In our paper, we derive a framework to assess the semantic FAIRness of dataspaces, which we directly apply to investigate the FAIRness of three mature initiatives. Our framework provides a basis for rating further initiatives and third-party developments that also deal with the development of dataspaces. Research and innovation projects such as MobiSpaces [58], Green.Dat.AI29 or Flex4Res 30 , which aim to use dataspace technology, have not yet been assessed. The same applies to projects funded by companies or through private programs that already provide the essential elements for setting up dataspaces, e.g., Solid [59].
Theoretical Contribution. This paper synthesizes the existing literature to provide a comprehensive overview of the evolution and current state of semantics in dataspaces. By extending the FAIR Principles, we propose a framework with tangible requirements that provide semantic clarity and interoperability. Accordingly, our paper establishes a foundation for evaluating future dataspace approaches based on the extended FAIR Principles, promoting a structured and objective assessment process.
Practical Implication. Connector implementations can leverage our extended FAIR framework to derive the requirements that a dataspace should fulfill. Vocabularies could be further developed and concretized specifically for the requirements found to cover potential demands.
Limitations. The analysis is limited to a specific number of dataspace developments, potentially missing emerging trends and niche innovations. Insight into the development and operational intricacies of these dataspaces is limited, relying on publicly available information and academic publications. Standardization processes within dataspaces often have a pay-asyou-go approach, which introduces variability in the implementation and adherence to the proposed FAIR Principles, affecting the generalizability of our findings.
Our literature review suggests that semantic approaches in theory and practice can guide the FAIRness of dataspaces. Semantic tools focus on standardization, establishing a uniform understanding through shared and standardized vocabularies. They support all FAIR Data principles and contribute to improvements. Utilizing standardized identifiers, shared ontologies, and rich contextual metadata, semantics enhance the individual aspects of the FAIR Principles and create a more interconnected and efficient data ecosystem. Current dataspace initiatives demonstrate similar approaches, with most FAIR Principles already guaranteed by semantics. However, there are additional approaches available, especially in automation such as our identified requirements pipeline, integration, and enrichment. Future directions could see dataspaces evolving to facilitate automatic data exchange and analysis, thereby improving the efficiency of data use. Capable and reliable integration systems can provide a larger data basis that can be further enhanced through enrichment. This includes further research into the development of advanced, dynamic ontologies and the automation of ontology matching for the seamless integration of different data sources. To ensure scalability, the continuous optimization of graph databases and the use of parallel processing to efficiently manage large amounts of data are essential. In addition, developing secure, ethical semantic technologies to protect privacy and promote responsible data use is a necessity if stakeholders are to share their data.
This publication is based upon work from COST Action DKG (CA19134), supported by COST (European Cooperation in Science and Technology). This work has been partially funded by the German Federal Ministry for Economic Affairs and Climate Action (BMWK) through the Antrieb 4.0 project (Grant No. 13IK015B), by the European Union's funded Projects MobiSpaces (Grant agreement no 101070279), Green.DAT.AI (Grant agreement no 101070416), AgriDataValue (Grant agreement no 101086461) and by the FAIR Data Spaces project of the German Federal Ministry of Education and Research (BMBF) under the grant number FAIRDS05.
From the literature, we extract the semantic approaches that are applicable in dataspaces. We code and list the ten most frequently mentioned ones, assign them to the underlying FAIR Principle according to their intended use and create from this the requirements profile for dataspaces, as shown in Table 1. Then, based on the analysis of architecture documents and code repositories, we compare the extent to which the semantics in the three currently most popular dataspace initiatives ensure FAIRness according to these criteria.