DIPAS Analytics – AI-powered Analysis in Citizen Participation

Content

Until now, the analysis of large volumes of textual citizen contributions has been largely manual, methodologically non-standardised, and has required significant human resources. DIPAS Analytics is a modular, AI-powered NLP toolbox that efficiently pre-structures text contributions, recognises patterns, aggregates content, and makes it suitable for spatial and statistical analysis. This enables those responsible for the process to gain a clear overview more quickly, identify key areas, link them together and present results consistently. The toolbox is currently being rolled out in Hamburg and will be released as open source in 2026.

The Key Learnings

AI provides support – but people must make the decisions

AI provides structure and suggestions, but no ready-made assessments or analyses. It is important that project managers are able to intervene, make corrections or reject suggestions at any time.

Without proper technical preparation, there can be no good AI

The quality of the results depends largely on the quality of the training datasets provided, on manually created gold standards, and on clearly defined categories, good descriptive text and contextual information.

Transparency is not an extra, but a requirement

AI results must be clearly labelled, explainable and traceable. This is the only way to build trust among users – and ensure long-term usability in an administrative context.

AI projects are organisational projects, not purely IT projects

As well as technology, it requires time, expertise, training, governance and ethics. These aspects must be taken into account from the outset, and appropriate resources must be allocated for them.

Explore More Academy Posts

Contact

Antonie Casper

“Community Participation”, Department of Urban Development and Housing, “Urban Workshop and Participation Processes”, Office for Regional Planning and Urban Development of the Free and Hanseatic City of Hamburg

digitalstrategien@sk.hamburg.de