The text on this page was automatically translated and hence may differ from the original. No rights can be derived from this translation.
Dialogic, together with Leiden University, the University of Amsterdam, and Utrecht University, investigated **how** and **how often** the various exemption grounds of the Government Information (Public Access) Act (Woo) are used to withhold documents or parts thereof from the public. This research aims to provide insights into patterns and differences in application and interpretation, facilitating informed discussions on transparency and careful consideration of interests based on facts.
This research is part of an Action Point of the Open Government Action Plan. Within the context of the Open Government Partnership (OGP) – a global movement with over 75 participating countries since 2011 – the Netherlands develops an action plan once every two years. The objective is for government organisations, civil society, and citizens to collaborate towards a more transparent government: one that is more effective and empowers citizens and civil society to collectively address societal issues. The action plan outlines the ambitions of the government and other stakeholders.
We have mapped the use of exemption grounds both quantitatively and qualitatively. Dialogic's expertise in applying AI is highlighted in this process. We have developed a methodology to identify patterns in the use of exemption grounds in Woo decisions. The research analysed 6,746 Woo case files, including 8,774 decision documents fragmented for analysis. A language model determined whether an exemption ground was cited in each fragment and identified the specific grounds mentioned. Only fragments where a ground was detected were considered relevant. The University of Amsterdam integrated this data into WooZM (the Woo search engine), allowing users to search through these fragments to understand how each ground is justified in different fragments.
To identify patterns on a large scale, a "map" was created. Relevant fragments were colour-coded based on the mentioned exemption ground and clustered according to their textual content. Fragments with similar text were placed close to each other, providing a quick overview of the content of explanations without the need to manually read through all fragments.
The research demonstrates that language models can effectively recognise exemption grounds in Woo decisions automatically. Furthermore, it reveals that justifications for the same exemption grounds often exhibit strong substantive similarities, allowing clear patterns and recurring formulations to be identified through clustering. Certain exemption grounds, such as the protection of personal policy views, business and manufacturing data, and personal privacy, frequently appear in Woo decisions. Additionally, different themes can be observed within the same exemption ground, indicating that a ground is utilised for various purposes. At a structural level, decisions often follow a similar format, with fixed components for assessing, weighing, and justifying exemption grounds.
Read the report for further insights into the methodology and results. Explore the interactive "map".
