Extracting Arguments and Motivations In Large-Scale Group Decision-Making: An Approach Using Large Language Models and Knowledge Graphs

Large-Scale Group Decision-Making (LSGDM) presents complex challenges related to scale, opinion diversity, and information overload. This paper explores two complementary approaches to address key aspects of LSGDM in online environments. The first focuses on argument mining from unstructured discursive data, using a modular pipeline powered by Large Language Models (LLMs) to extract, classify, and cluster argumentative content. The second targets the extraction and categorization of user motivations in structured discussions, employing psychological theories—particularly Max-Neef’s Model of Human Scale Development—to identify underlying human needs. Both approaches model discourse using knowledge graph structures, providing interpretable representations of public reasoning. Reddit served as a shared data source due to its volume, diversity, and accessibility, allowing the development and validation of techniques transferable to more formal deliberative platforms. Results show both systems can uncover dominant argumentative patterns and synthesize recurring motivational drivers, highlighting the multidimensional nature of online deliberation. The integration of argument structure and motivational layers provides novel insights into collective reasoning, supporting more transparent, inclusive, and context-aware decision-making. These contributions establish a computational foundation for discourse analysis in participatory contexts and offer practical tools for civic engagement and e-democracy initiatives.