Moderate Code

We implement a reproducible pipeline to examine the emotional characteristics of large language models in polarized online discourse. Using a large-scale dataset of posts from Twitter and Reddit covering climate change, the 2016 and 2020 U.S. presidential elections, and the QAnon conspiracy theory, we prompt seven state-of-the-art LLMs to generate responses under a unified prompting and decoding configuration. Both model-generated and human texts are processed through the same emotion analysis module to quantify emotional polarity and intensity. The resulting distributions are statistically compared across topics and platforms to assess systematic differences between LLM and human emotional expression.