Improving digital citizen participation through AI-evaluation tools I Claudius Lieven
Claudius Lieven is head of the Stadtwerkstatt, a staff-unit for citizen participation at the Ministry for Urban Development and Housing in Hamburg, Germany. He is a political scientist and led the development of the Digital Participation System (DIPAS). He is currently working on the development of digital tools for urban development as part of the Connected Urban Twins project. He is a member of the board of trustees of the Federal Association for Housing and Urban Development and the Alliance for Diverse Democracy.
The talk was given at the Urban Digital Twins for a Sustainable Transformation of Cities conference at City Science Lab, HafenCity University Hamburg, Germany.
More Information: https://www.citysciencelab.hamburg/projects/twin-conference-2025
thanks a lot Rico i’m very happy to be able to speak to you i’m uh not from science not a scientist but a practitioner so it is a report from the workshop translates to urban workshop and here we are at the end of the presentation it’s a bit too fast so well how that here okay um one sentence to the Deepass system deepass is digital participation system we began with that some years ago it has various modules online participation digital participation on site with touchts just now today we have the inauguration of a a storytelling tool that was developed here in the CSL we had integrated it into deepass and now we have digital storytelling for urban development projects just 3 kilometers from here right now today uh it it’s happening and we have a deepest navigator that gives us the overview over all the processes we carry out with that because since 2016 meanwhile we are nearly at 130 uh processes and we have collected tens of thousands of contributions last year over 10,000 contributions to together with um 5,000 commentaries on contributions and this is all integrated into one uh big system and in one database and Um a problem that arises is that um if you have these huge loads of feedback of the citizens we are absolutely happy that is on urban planning it’s uh they give us really their knowledge um and respond to our questions but how do you evaluate that it is for the uh local authorities who are carrying out it’s this is a district of Altona with 3,000 response this is huge so we have statistical analyzis tools this is in the cozy system it’s as well a part of our urban digital twin so to say we have an API we exchange data with the system and we have a statistical analysis the bigger the green dots the more response like here we have 107 positive evaluations on this contribution and but we as well see for instance red dots and that means we have negative evaluations and here we have a conflict line it is something we didn’t ask for but it showed up in the data because we asked for bicycle traffic and this is a a road problem but anyway you can reduce complexity with this statistical analysis quite good but it stems only the analysis from the numbers and not from the text the text the statistic tool cannot read so we asked ourself how can we improve the evaluation of this huge amounts of information and we came to the point okay let’s build an AI system uh but easy said and not so easy done and um so we formed a project and it was u we we are steering it my unit theat and at the moment I’m speaking on behalf of Matias Lensinski he’s the chief of this sub project so to say but at the moment on Aruba and in the Caribbe so I’m doing it and we have the LGV our agency for geo information in uh um uh yeah geo information and we have the HCU which here did specially the front- end design and we have scientific partners because all the argument analysis the text analysis that was developed in other project we exchanged with them um data for years and they carried out research and then we took back their u results and formed um well our um language processing model But um we’ll have two um main functionalities we have an analytics life dashboard this is for the responsible persons of the proceedings because they run for some weeks and they need to look every day into it and how is it going and as well a list of posts that may need to be moderated mostly the problem that occurs that people write into the contributions personal data and we were completely free of personal data that’s why we can exchange data otherwise under European law that would be difficult but the more interesting part is the insights tool because this is after the process really an do the report and uh synthesize the report we asked our the colleagues who are currently doing the evaluations what do you do in the process and they said we first need to understand what have the people written then we need to connect these items of information we need to cluster them we need to bring topics together and then we analyze it spatially so we have a text evaluation and of course we want to represent the information into the space and then we need to communicate that to the authorities to the planning boards and all these And this is done via the insights tool i don’t go into details but what we now can do and see is that we uh analyze patterns of response patterns of impression patterns of um up to the feelings of the people which they express in the text and we can connect that to spatial patterns so for instance if we ask for bicycle traffic we can overlay that with a bicycle grid in the city but here we see for instance the the overlay is not perfect and maybe it is another topic that is more important here it’s maybe the schools uh around which um the uh situation the security situation needs to needs to be improved we did we have a pre-processing pipeline i do it very short we have the API the whole system has an API and scientific uh institutions can get the data from via this API and work with our data and we end with sentences because in one contribution often there are some major positions some topics and so this is all sliced and this is a pre-processing pipeline but the main thing is this main processing pipeline because we have a statement extraction a label classification category assignment cluster identification identification and lastly a title generation for the clusters and the aspects various um LLM tools come into play here and we train them for our specific tools and they are all open source i cannot go into the details here uh we’ll have a conference in September and if you are interested you’re invited and they will go very deep into the details for that just to show this we are nearly at the end now in 2025 we’ll have an open source release of the entire um add-on part of DPAS and we have the special uh open-source release for the NLP um LLM tools we just ended with a front end development of the insights tool some things are still to fix then we’ll have um qualification concept to do and to set up uh the training of the all the experts in the districts and anywhere to use the system because it is a little bit complex and um on the 15th of September we are planning to have a launch conference because the system the all the back end is operational parts of the front ends as well as are operational we begin now with internal testing and um yeah in September we are going to roll it out we have meanwhile 10 cities using and institutions using DEPAS in Germany and uh because they can as well access this system via the API they have the API as well so we can replicate that very easily uh it works in German we must say it’s the the LLMs are trained in German so we cannot uh directly put it into other languages it would be a step ahead in the next years thank you for your attention