From now, Dragonfly Data Science will house Croissant, an analysis tool that streamlines the process of analysing public submissions.
Croissant was originally developed by the Ministry of Environment (MfE), the Department of Internal Affairs, and Dragonfly Data Science. However, MfE no longer has the capacity to manage Croissant, and without regular oversight and updates, the tool is unable to deliver on its full potential.
To avoid this, Dragonfly Data Science will now host the tool, ensuring that Croissant will continue to benefit the public service and the general public.
“Croissant is a product that we’re very excited to take onboard to refine and develop further,” says Dragonfly’s Director of Data Science, Dr Finlay Thompson. “Public submissions are a core part of the democratic process. When large numbers of people take the time to engage with a bill or public consultation, it is important their views can be properly considered. “Croissant helps agencies review and understand that feedback in a way that is more manageable, more consistent, and better suited to the scale of modern consultation.”
What is Croissant?
Before laws are introduced, or policies changed, the public may be invited to submit their views on proposals through varying channels, including templated submissions, email, or handwritten documents.
A consultation can produce tens, or even hundreds, of thousands of responses, all of which need to be considered. Croissant is an analysis tool designed to streamline and standardise the analysis of these submissions.
How does it work?
Croissant tags each submission with keywords that relate to its content. From there, submissions are sorted and analysed. Tags also identify submissions made from specific groups (such as companies, iwi, or regions) to ensure they are addressed appropriately.
The tool shows the status of all of your consultations, the submission themselves, and how much of the text has been tagged.
There are three different modes that Croissant works in: manual, unsupervised, and supervised.
The manual mode allows users to upload a taxonomy and manually tag sections of written text that align with categories within that taxonomy (for example, submissions that support the proposal versus those that do not). Croissant acts as an enormous set of digital highlighters, assigning a different ‘colour’ to each tag to visually represent the results. It then records and counts the number of submissions under each tag.
In an unsupervised mode, the machine learning module identifies groups of submissions that contain similar ideas. With no human intervention required, this mode provides an overview of a large body of text and creates an interactive map of all submissions, grouping similar ones together. The results are then used to help develop tags that are appropriate for all the submission text.
In a supervised learning mode, the module is taught to associate manually applied tags with specific text under review. It then automatically applies those same tags to the remaining untagged text.
Why use Croissant?
Croissant allows people to quickly and easily understand written data, improving the process of how public submissions are organised, freeing up resources to deliver for the public.
Traditional methods of analysing large bodies of written information are extremely time-consuming. MfE estimates that Croissant reduces the time needed to manually tag submissions by up to 90 percent.
Lastly, security measures built into Croissant help protect the submitter’s personal information.