Published September 25, 2023 | Version 1
Technical note Open

Datasheets for Digital Cultural Heritage Datasets

  • 1. CARARE, Dublin, Ireland
  • 2. KB, National Library of the Netherlands, The Hague, Netherlands
  • 3. Department of Classical and Italian Philology, University of Bologna, Italy
  • 4. School of Social Sciences and Humanities, NOVA University of Lisbon, Lisbon, Portugal
  • 5. Europeana Foundation, The Hague, Netherlands
  • 6. Staatsbibliothek zu Berlin – Berlin State Library, Berlin, Germany
  • 7. School of Information and Communication Studies, University College Dublin, Dublin, Ireland
  • 8. Hugging Face, Glasgow, Scotland


As the collections as data paradigm is gaining momentum, fueled by powerful advancements in machine learning and data mining technologies, the institutions managing digital cultural heritage collections urgently need to learn how to provide meaningful information about their collections as data, i.e. not on the item level, but on the collection level. Datasheets just do that.

Similar to instruction leaflets, they document the context and content of datasets needed for re-using such datasets and enable transparency and accountability. They can mitigate unwanted biases in machine learning models, facilitate reproducibility of machine learning results, and help researchers to choose the right dataset. They open up a space for negotiation and facilitate interdisciplinary communication between cultural heritage practitioners, researchers and technical experts. While the concept of datasheets was introduced to the machine learning community by Gebru et al (Datasheets for Datasets, and Pushkarna et al (Data Cards,, it still lacks adaptation to the requirements of the European cultural heritage field. Cultural heritage datasets differ from contemporary, industrial datasets in many ways: they are heterogeneous with respect to the time period covered, the place or regions incorporated in them, or the cultural contexts in which they have to be located. They may contain sensitive content, e.g. in the case of a collection of sources from former colonies, and thus may require ethical questioning. Metrics—highly estimated by machine learners—often are not helpful to describe such datasets. Because cultural heritage datasets grow over time, the data sheets need to be adaptable, taking enlarged datasets and changing uses into perspective. The complexity of such data is often underestimated by computer scientists, which impedes the kind of scientific negotiations of meaning and interpretive transfers which datasheets aim to facilitate.

Resulting out of a Europeana Working Group, we present here a datasheet template for digital cultural heritage datasets. We explicitly understand this template as a proposal, intended for gaining experience and to collect feedback. It is important to stress that the template should be thought of as modular: it is up to those filling in the form to decide which questions should be answered and which questions could be ignored.


Template Datasheet for Digital Cultural Heritage Datasets - V1.pdf

Files (144.8 kB)