OATH-Frames: Characterizing Online Attitudes Towards Homelessness with LLM Assistants
Public attitudes towards key societal issues (e.g., homelessness), are of immense value in policy and reform efforts, yet nuanced and challenging to understand at scale. This project introduces a mechanism to understand these attitudes at a high-level and at a large scale, capturing critiques, responses and perceptions of online attitudes towards homelessness. We study the effectiveness of our method by comparing them to known events surrounding homelessness, and find that our methods coincide with reported attitudes towards government policies on homelessness. We hope that this work can lead to effective messaging about such critical societal issues that can lead to positive changes in opinions and awareness.
Goals
To leverage advances in AI to understand what are the public attitudes towards homelessness as expressed in online discourse at a large scale.
Methods
We first build a framing typology, called Online Attitudes Towards Homelessness (OATH). This includes nine hierarchical frames, spanning those for carefully curated attitudes on critiques, responses and perceptions on the issue. We propose a human-in-the-loop mechanism, which uses large language models to assist with the prediction of these attitudes on a large collection of about 3M posts from X. We release annotations with varying degrees of assistance from language models, with immense benefits in scaling: 6.5x speedup in annotation time compared to humans, while still maintaining a high degree of accuracy.
Swabha Swayamdipta
Eric Rice
Jaspreet Ranjit
Kang, J. 2024, October 13. How a USC student helped build an AI system that helps gage public sentiment towards homelessness. Spectrum 1 News, https://spectrumnews1.com/ca/southern-california/technology/2024/10/13/ai-usc-advocacy-homelessness-la
Outstanding Paper Award at Empirical Methods of Natural Language Processing Conference, 2024.
