Large language models to complement the services offered by the city
The City of Aachen's website has services such as the the city of Aachen has valuable opportunities for citizens to citizens to participate in shaping the city or to voice their concerns. their concerns.
In order to be able to process such feedback from citizens means, it makes sense for citizens to look for the responsible office and submit their concerns or feedback there.
LLMs (Large Language Models) offer the opportunity to make the city's make the city's services easier for citizens to find. Some services are already included in the knowledge base of publicly accessible LLMs such as ChatGPT (e.g. the 4o model). For example, the input that something is growing in the cycle path in Aachen leads to a response that links directly to the defect report.
Native integration on the website can make the city's services city's services even easier to find.
HT
Henrik Thillmann
Doktorand im Software Engineering an der RWTH Aachen
Content Blocks
Expand feedback options
Even with such a supplement, I still have to find the right contact person and decide who to contact. It may even be necessary to create an account.
Ideas, criticism or praise from citizens that do not exceed this effort threshold will remain unused.
One possibility is to use a free text field, for example on the the homepage of the website to create a low-threshold offer for feedback. create.
While for non-intelligent algorithms such a such a data set is worthless or requires a great deal of manual effort to process, the use of processing, the use of LLMs and RAG (Retrieval Augmented Generation) allows knowledge to be to extract knowledge from such a data set.
The free text entries can be assigned to the city's existing city's existing services or processed by an LLM for use by city departments. usable by an LLM.
In principle, other text input channels can also be used can also be used: Apps, messengers. Voice-to-text processes can also be used for audio sources, which further reduce the barrier to use, especially on mobile devices. devices in particular.
The integration on the website would probably initially require the would initially require the least development effort.
One example of such low-threshold feedback is that the information page of the outdoor pool did not indicate how long the diving the diving tower was affected by the renovation work. That is information for which I don't necessarily have to find out the responsible contact person, or even bother them with it. And such knowledge is then not available to the content managers for the website.
The extensive use of existing technologies and infrastructure technologies and infrastructure reduces the implementation risk and the development development effort.
Existing LLMs either already have enough internal knowledge knowledge to be able to recommend and link services of the city, so that no separate data structure (knowledge graph etc.) needs to be created, or they can process text data, such as from the website.
Possible questions that can then be posed to a data set with citizen feedback and could be answered by an LLM are:
What features do users of the city's website city's website most? What ideas have been put forward?
Technologische Risiken und Umsetzungsaufwand
Die weitestgehende Verwendung von bereits existierenden Technologien und Infrastruktur senkt das Umsetzungsrisiko und den Entwicklungsaufwand.
Bestehende LLMs besitzen entweder bereits genug internes Wissen, um Services der Stadt empfehlen und verlinken zu können, so dass keine separate Datenstruktur (Knowledge Graph etc.) erstellt werden muss, oder können Textdaten, wie von der Website verarbeiten.
Conditions of participation
You can view the conditions of participation under "Link".