As a spin out from TU Darmstadt, SUMMETIX is committed to supporting cutting edge AI and LLM research.
Recently, we supported research led by the Dialogue Systems Group at Ulm University, resulting in the paper “Automatic Generation of Structured Domain Knowledge for Dialogue-based XAI Systems” presented at IWSDS’25.
In their study, the researchers explore how a pipeline designed to automatically create argumentation trees for debates (persuasive, argumentative dialogues) could be adapted to model domain knowledge for explanations (explanatory dialogues). With the domain knowledge, the explanations provided by methods from the field of XAI (eXplainable AI) can be enhanced for non-expert users.
Normally, when building a system that guide users utilizing information in a complex knowledge data base, a lot of manual work is required to build the knowledge structure.
By employing SUMMETIX’s Classify API in their pipeline, the researchers could automate the generation of a knowledge structure, where arguments are automatically structured into a tree.
In the scenario of applying for a bank loan, the fully automatic instantiation of the pipeline performed well compared to a much more time-consuming manual set up.
With a solid set of documents to extract the information from, the pipeline promises to be successful when creating XAI systems with domain knowledge relying on very large qualitative data sets.
