Challenges to Enforce Data Quality in Data Spaces Claudia P. Ayala1 , Besim Bilalli1 , Cristina Gómez1 , Jose-Norberto Mazón2,∗ and Oscar Romero1 1 Universitat Politècnica de Catalunya, BarcelonaTech 2 Universitat d’Alacant / Universidad de Alicante Abstract Data Spaces must preserve sovereignty and privacy while ensuring FAIR (Findable, Accessible, Interoperable and Reusable) principles. To do so, policy-based strategies have to be developed in order to describe the agreements reached in the Data Space. In this context, two open questions arise: how to define the right Data Space policies, as well as, how to enforce (and monitor) them. Despite the efforts towards defining and enforcing data access and usage policies, there is no solution to operationalize the enforcement of those considering data quality dimensions. However, data quality is becoming a hot topic due to the surge of federated learning and alternative analytical techniques, which require all providers to guarantee a data quality threshold in order to learn robust models. Currently, we have means to describe policies related to data quality rules (e.g., by combining standards such as ODRL and standard vocabularies) but we are missing means to elicit these policies from data providers and enforce them while preserving the data sovereignty. In this paper, we discuss the challenges and open questions that must be addressed in order to operationalize (and eventually, automate) data quality in Data Spaces, which span from requirements elicitation to data validation. Keywords Data Spaces, Data Quality, Data Validation, Federated Data Management, Data Sharing 1. Introduction DQ) at the Data Space federated layer (i.e., at the federated -unique- view of the data ecosystem) can be enforced at Data Spaces are federated ecosystems in which data the providers’ data assets regardless of their heterogeneity providers and consumers share data while preserving data and preserving data ownership and privacy. Note that this sovereignty and privacy. Currently, the Data Mesh archi- problem has been easily tackled in centralized environments tecture [1] is at the core of current technological solutions, by having a central authority extracting, transforming and since it provides a domain-decentralized paradigm that suits preparing data for analysis. However, this is not possible in the Data Space requirements [2]. Relevantly, the Data Mesh settings where data is not meant to be shared raw. For ex- defines the Data Product concept, which provides a product- ample, the minimum number of instances and the variances oriented view of the providers’ data assets. In short, the data of key attributes might be set as DQ criteria for all data product is a node that encapsulates three structural compo- providers and should be automatically and locally validated nents required to function: code for enforcing policies (i.e., by executing a software service (specific for the provider the Data Space agreements), data (and its metadata) and infrastructure) provided by the Data Space services catalog. infrastructure [3]. By definition, the providers’ data assets The result of the service execution should be communicated can be heterogeneous both in the infrastructure used and to the Data Space. To our knowledge, there is no architec- the data provided (in format and semantics). ture, framework or solution tackling this problem, despite Behind the idea of Data Spaces is the objective of extract- the myriad of standards and definitions blooming around ing value from data sharing. This can be achieved in many the Data Space concept (e.g., [7, 8]). ways, but data analysis arises as prominent means to achieve We focus on how to validate DQ agreements in the Data so, either by means of descriptive analysis (e.g., dashboard- Space and discuss the open challenges to make DQ happen ing and OLAP) or predictive analysis (e.g., learning models). in Data Spaces to enact trustworthy federated learning. However, how to achieve data analysis in federated envi- ronments is an open challenge, and federated learning [4] is currently the most widespread privacy-aware data analy- 2. Challenges and Vision sis technique. Many efforts have been devoted to develop robust federated learning but little attention has been paid Data Spaces require a governance model for specifying DQ to the role of data. Yet, the impact of the data quality (DQ) agreements that stakeholders must adhere to in order to par- from each provider on federated models learnt is huge [5, 6]. ticipate. Importantly, this governance model must also spec- One of the biggest open problems in Data Spaces not ify DQ needs agreed among data consumers and providers properly tackled is how the agreements reached (e.g., on when developing specific uses cases. Therefore, our view is that the governance model for Data Spaces should dis- DOLAP 2025: 27th International Workshop on Design, Optimization, Lan- tinguish two levels: 1) a Data Space level for agreements guages and Analytical Processing of Big Data, co-located with EDBT/ICDT among stakeholders of the Data Space authority from data 2025, March 25, 2025, Barcelona, Spain ∗ Corresponding author. regulations and strategic issues, and 2) a use case level for Envelope-Open claudia.ayala@upc.edu (C. P. Ayala); besim.bilalli@upc.edu agreements among data providers and consumers to build (B. Bilalli); cristina.gomez@upc.edu (C. Gómez); jnmazon@ua.es specific Data Products. Based on this view and to facilitate (J. Mazón); oscar.romero@upc.edu (O. Romero) the discussion, we propose a visionary framework with a GLOBE https://futur.upc.edu/ClaudiaPatriciaAyalaMartinez (C. P. Ayala); process for the Data Space and use case levels (see Fig. 1). https://futur.upc.edu/BesimBilalli (B. Bilalli); https://futur.upc.edu/CristinaGomezSeoane (C. Gómez); Our framework follows the Open Data Product specifica- https://s.ua.es/_MuH (J. Mazón); tion [9], thus splitting each process into two parts: one https://futur.upc.edu/OscarRomeroMoral (O. Romero) declarative, at a higher-level of abstraction specifying what Orcid 0000-0002-6262-3698 (C. P. Ayala); 0000-0002-0575-2389 (B. Bilalli); (analysis phase), and another one at a lower-level specifying 0000-0002-3872-0439 (C. Gómez); 0000-0001-7924-0880 (J. Mazón); how (design and implementation phases). The declarative 0000-0001-6350-8328 (O. Romero) © 2025 Copyright for this paper by its authors. Use permitted under Creative Commons License part defines the DQ dimensions and intended level. The ex- Attribution 4.0 International (CC BY 4.0). CEUR ceur-ws.org Workshop ISSN 1613-0073 Proceedings Figure 1: Visionary framework for considering DQ requirements in Data Spaces. ecutable part contains the machine-readable “as code” rules, We propose to use a rule language with well-defined seman- provided as a service, to validate DQ dimensions. Next, we tics (e.g., ODRL), to formalize DQ rules. Several challenges describe both processes and their main challenges. need to be tackled when performing this transformation: 1) DQ Requirements Engineering for Data Spaces. the identification of relevant and suitable stakeholders with Requirements engineering (RE) for complex systems in the specific knowledge for performing this activity in both open and dynamic environments that extend beyond a sin- levels; 2) the definition of specific catalogues with reusable gle organization is widely recognized as a challenging en- transformation patterns for translating DQ requirements deavor [10, 11]. This is particularly true in the context of into rules, preserving their semantics; 3) the definition of the Data Spaces, where the elicitation and management of re- artifacts needed (e.g., specialized metamodels or new ODRL quirements must reconcile diverse perspectives, including profiles), for automating the extraction and customization the strategic business vision, governance, compliance with of DQ rules to the specific domain and level. laws and regulations, infrastructure, scalability demands, Implementation available as a Service of DQ Rules. and DQ considerations. Our visionary framework proposes The inherent heterogeneity of providers in the context applying RE practices to elicit, specify, and manage the Data of Data Spaces renders the process of translating formal Space requirements. We advocate for the development and DQ rules into executable services a significant challenge. use of a Catalogue of DQ Requirements at two levels: the The main goal of this activity is to avoid building and main- Data Space level and the use case level. These catalogs pro- taining custom solutions that are tightly coupled to specific mote knowledge sharing and requirements reuse, building a execution environments or platforms. To address this, we robust repository of experiences and best practices. The pro- propose an agnostic solution that leverages best practices posed process is aimed to: 1) Ensure a common understand- from software engineering, such as containerized solutions, ing of DQ dimensions by considering established standards; ensuring portability, scalability, and interoperability. How- 2) Facilitate the elicitation of diverse DQ requirements from ever, the intrinsic characteristics of Data Spaces introduce diverse stakeholders to enable effective data sharing; 3) Sup- several challenges that must be addressed: 1) dealing with port the structured specification and management of DQ heterogeneity at the infrastructure level by abstracting the requirements to ensure compliance and alignment between differences while ensuring consistent performance and secu- the Data Space and use case levels for their subsequent oper- rity across environments; 2) allowing for dynamic and feder- ationalization; and 4) Address trade-offs between conflicting ated execution across multiple distributed nodes, ensuring DQ requirements. This approach aims to bridge the gap real-time validation without requiring data centralization. between diverse stakeholder perspectives and the technical As conclusion, there is a need for further research to requirements for robust DQ management in Data Spaces. enact DQ in Data Spaces, a must for qualitative federated Extraction and Customization of DQ Rules. The data analysis. In this sense, we have discussed a visionary complexity of DQ requirements and their textual or semi- framework, its main phases and challenges to be tackled. structured formalization make their direct operationaliza- tion challenging. With the aim of making DQ requirements executable in an operational environment, our visionary Acknowledgments framework proposes to transform, in a semi-automated way This work has been partially supported by the EU- and using specific catalogues for supporting this transfor- HORIZON program under GA.101135513 (CYCLOPS) and mation, DQ requirements (at Data Space and use case levels) by CIAICO/2022/019 project from Generalitat Valenciana. into formalized DQ rules that may be easily implemented. References [1] A. Goedegebuure, I. Kumara, S. Driessen, W.-J. Van Den Heuvel, G. Monsieur, D. A. Tamburri, D. D. Nucci, Data mesh: a systematic gray literature review, ACM Computing Surveys 57 (2024) 1–36. [2] M. Bacco, A. Kocian, S. Chessa, A. Crivello, P. Barsoc- chi, What are data spaces? systematic survey and future outlook, Data in Brief 57 (2024) 110969. [3] Z. Dehghani, Data Mesh: Delivering Data-driven Value at Scale, O’Reilly, 2022. [4] B. McMahan, E. Moore, D. Ramage, S. Hampson, B. A. y Arcas, Communication-efficient learning of deep net- works from decentralized data, in: A. Singh, X. J. Zhu (Eds.), Proceedings of the 20th International Confer- ence on Artificial Intelligence and Statistics, AISTATS 2017, 20-22 April 2017, Fort Lauderdale, FL, USA, vol- ume 54 of Proceedings of Machine Learning Research, PMLR, 2017, pp. 1273–1282. [5] T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, V. Smith, Federated optimization in heterogeneous networks, in: I. S. Dhillon, D. S. Papailiopoulos, V. Sze (Eds.), Proceedings of the Third Conference on Ma- chine Learning and Systems, MLSys 2020, Austin, TX, USA, March 2-4, 2020, mlsys.org, 2020. [6] A. Jain, H. Patel, L. Nagalapatti, N. Gupta, S. Mehta, S. Guttula, S. Mujumdar, S. Afzal, R. Sharma Mittal, V. Munigala, Overview and importance of data qual- ity for machine learning tasks, in: Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, Association for Computing Machinery, 2020, p. 3561–3562. [7] Fiware for data spaces: Position paper, https://www.fiware.org/wp-content/uploads/ FF_PositionPaper_FIWARE4DataSpaces.pdf, 2024. Accessed: 2024-12-20. [8] International data spaces association, https://internationaldataspaces.org/why/ international-standards/, 2024. Accessed: 2024- 12-20. [9] Data product specification, https://opendataproducts. org/v3.1/#optional-attributes-and-elements, 2024. Ac- cessed: 2024-12-20. [10] P. Malcher, E. Silva, D. Viana, R. P. dos Santos, What do we know about requirements management in software ecosystems?, Requir. Eng. 28 (2023) 567–593. [11] P. Hagenhoff, S. Biehs, F. Möller, B. Otto, Design- ing a reference architecture for collaborative condi- tion monitoring data spaces: Design requirements and views, in: M. Mandviwalla, M. Söllner, T. Tuu- nanen (Eds.), Design Science Research for a Resilient Future - 19th International Conference on Design Sci- ence Research in Information Systems and Technology, DESRIST 2024, Trollhättan, Sweden, June 3-5, 2024, Proceedings, volume 14621 of Lecture Notes in Com- puter Science, Springer, 2024, pp. 355–369.