Deciphering Conversational Networks: Stance Detection via Hypergraphs and LLMs

Understanding the structural and linguistic properties of conversational data in social media is crucial for extracting meaningful insights to understand opinion dynamics, (mis-)information spreading, and the evolution of harmful behavior. Current state-of-the-art mathematical frameworks, such as hypergraphs and linguistic tools, such as large language models (LLMs), offer robust methodologies for modeling high-order group interactions and unprecedented capabilities for dealing with natural language-related tasks. In this study, we propose an innovative approach that blends these worlds by abstracting conversational networks via hypergraphs and analyzing their dynamics through LLMs. Our aim is to enhance the stance detection task by incorporating the high-order interactions naturally embedded within a conversation, thereby enriching the contextual understanding of LLMs regarding the intricate human dynamics underlying social media data.