Dutch FOIA Documents Turned Into a Standardized Research Dataset
A new FAIR collection of Dutch Freedom of Information Act documents gives researchers access to an unusually large body of government records in a form that can actually be studied at scale. The dataset contains just over two million pages, gathered from publicly released materials and organized so the documents, their metadata, and available links to source PDFs can be used more reliably across disciplines.
The project addresses a familiar problem in public-sector data work: the information exists, but the structure does not. Dutch citizens can request information from government bodies under the Freedom of Information Act, and the released files are often published across many organizational websites with inconsistent formatting, uneven metadata, and no shared standard for how the material should be labeled or stored. For researchers trying to analyze the documents computationally, that fragmentation is not a minor inconvenience. It determines whether the records can be searched, compared, text-mined, or linked to other collections at all.
Why Standardization Matters for Government Records Research
The value of this collection lies less in the novelty of the documents themselves than in the way they have been assembled. The team scraped supplier websites, standardized metadata, and then checked the resulting files for text and metadata quality. In a dataset this large, those steps are not administrative housekeeping; they decide whether the corpus is usable for natural language processing, document classification, information retrieval, and social or political science analysis.
A government disclosure archive can be dense with policy language, institutional correspondence, and procedural detail, but that richness disappears quickly when file naming conventions vary and metadata fields are incomplete or incompatible. Researchers working across departments, ministries, and publication platforms often spend more time cleaning data than analyzing it. A collection built around FAIR data principles changes that balance by making the documents findable, accessible, interoperable, and reusable from the outset.
The practical effect is straightforward. When document text, layout, and metadata are organized together, the corpus becomes more than a pile of scanned pages or PDFs. It becomes a structured research asset that can support reproducible work, cross-institutional comparisons, and downstream methods that depend on predictable file architecture.
Two Million Pages, One Repository, and a Better Starting Point
The dataset is available through the DANS data repository, where users can access the document text and layout, the accompanying metadata, and, where available, links to the original PDF files. Usage instructions and examples are included as well, which lowers the barrier for researchers who may not have the time or technical infrastructure to rebuild a document collection from scratch.
That access model matters because the volume is substantial. Just over two million pages creates analytical possibilities that are difficult to achieve with small, manually curated archives. At that scale, researchers can study language patterns across agencies, trace recurring topics in disclosure requests, examine publication habits over time, or evaluate how administrative language shifts across organizations. The size of the collection also makes it suitable for computer science applications that require large corpora, including machine learning workflows and document layout analysis.
The release does not claim to solve every challenge associated with public records research. Inconsistent upstream publishing practices still shape what enters the corpus, and the quality of any large document collection depends on the source material itself. Even so, the assembly process represented here reduces a major obstacle: fragmented, hard-to-use records have been transformed into a coherent dataset with standardized metadata and documented access.
One sentence captures the core advance plainly: the documents are no longer trapped on scattered websites.
What the Collection Enables Across Disciplines
Although the dataset has obvious relevance for computer science, its usefulness extends into social and political research as well. Freedom of Information Act documents are a window into how institutions respond to public requests, what kinds of information are released, and how government decision-making is described once it enters the public record. For scholars studying transparency, administration, or institutional communication, the corpus offers a way to move beyond anecdote and sample-size limitations.
The repository format also supports reproducibility, a frequent weak point in large-scale document studies. When metadata are standardized and the source material is documented in one place, other investigators can inspect the same corpus, apply different extraction methods, and compare results more fairly. That is particularly useful in computational research, where small differences in document preprocessing can produce large differences in output.
From a technical perspective, the layout information may be as important as the text itself. Document structure can influence how text is extracted, how sections are segmented, and how tables or references are interpreted. A corpus that preserves both textual and layout information gives analysts more flexibility and reduces the risk of stripping away context during preprocessing.
The Research Value of Public Access Done Well
The collection also shows how public disclosure systems can become research infrastructure when publication practices are harmonized. FOIA releases are often treated as one-off administrative outputs, but a coordinated dataset changes their role. Instead of isolated files hidden across institutional pages, the records can function as a durable resource for secondary analysis.
That shift has consequences beyond methodology. In practice, it lowers the cost of studying government transparency, creates a more dependable base for training and testing computational tools, and gives nontechnical users a better chance of finding documents relevant to their work. For policy analysts, archivists, and digital humanities researchers, the difference between a disorganized archive and a FAIR repository is the difference between browsing and evidence generation.
The collection’s limitations are structural rather than clinical, and they are worth keeping in view. It reflects the documents that were publicly available and accessible through scraping, so its completeness depends on how organizations chose to publish their records. Quality checks can standardize and validate what is present, but they cannot create source material that was never released or was posted in ways that resisted capture. Even with those constraints, the dataset offers a rare combination of scale, organization, and documented reuse support.
For researchers who work with public records, that combination is the real story. Two million pages are impressive, but the more important achievement is turning a scattered release environment into a dataset that can be searched, compared, and reused with much less friction.