OATH-Frames: Characterizing Online Attitudes Towards Homelessness with Language Model Assistants

Watch a research seminar presented by USC PhD student Jaspreet Ranjit exploring how large language models can assist in analyzing public attitudes towards homelessness on social media. Learn about the development of OATH-Frames, a novel hierarchical framing typology consisting of nine frames that capture critiques, responses, and perceptions related to homelessness. Discover how this framework enabled researchers to analyze 2.4 million Twitter posts with the help of LLMs, achieving a 6.5x speedup in annotation time while maintaining strong performance. Explore key findings about attitudes across different states, time periods and vulnerable populations, demonstrating the potential for using AI to understand complex social issues at scale. Gain insights into collaborative approaches between social science experts and language models for studying sensitive societal topics, with implications extending beyond homelessness research.