Understanding Vegan Discourse on Social Media by Leveraging LLMs

Abstract

Social media platforms amplify discussions on lifestyle choices like veganism, shaping public discourse that influence consumer behavior, ethical debates, and environmental policy. Understanding the dynamics of these discussions requires scalable and efficient methodologies. This paper presents a novel framework for analyzing vegan discourse on social media by integrating large language models (LLMs) with advanced clustering techniques. We introduce a task of classifying vegan discourse into key themes, using a dataset of 20,000 tweets. Our experiments demonstrate that leveraging LLMs with advanced clustering algorithms, specifically HDBSCAN, enhances semantic coherence in clustering. Furthermore, LLMs show potential as an unsupervised annotator, significantly reducing the need for manual labeling. These findings provide a scalable framework for automating social media text analysis, offering valuable insights into public discourse on veganism.