This repository contains the code and data for our paper:
OATH-Frames: Characterizing Online Attitudes Towards Homelessness with LLM Assistants
Jaspreet Ranjit, Brihi Joshi, Rebecca Dorn, Laura Petry, Olga Koumoundouros, Jayne Bottarini, Peichen Liu, Eric Rice, Swabha Swayamdipta
We introduce a novel framework to understand, synthesize and characterize large-scale public attitudes towards societal issues through a collaboration between social work experts and LLMs. Specifically, we introduce a framing typology: OATH-Frames, (Online Attitudes Towards Homelessness): nine hierarchical frames capturing public attitudes towards homelessness as expressed on Twitter. We provide three kinds of annotations for posts from Twitter: expert-only, LLM-assisted expert and predicted annotations from a multilabel classification model.
- Install the recommended dependencies via Anaconda
conda create -n oath python=3.9.12 conda activate oath conda install -c conda-forge pip # make sure pip is installed python -m pip install -r requirements.txt # make sure the packages are installed in the specific conda environment python -m pip install -e .
Please refer to Hugging Face for our released data
- Expert Annotations: oath-frames-expert-annotations
- Expert + GPT-4: oath-frames-expert-plus-gpt-annotations
- Analysis set (Model Predicted): oath-frames-model-predicted-annotations
- Train/Test/Eval Splits: oath-frames-flan-datasets
- NER predictions: oath-frames-analysis-ner
- PEH/Vulnerable population analysis (Section 4.3) in the paper: oath-frames-analysis-vulnerable-populations
Note: Posts labeled with 0
, []
, or do not have any labels are those that have been filtered out as irrelevant to our task. Please exclude these during analysis
Please refer to src/trainer_deepspeed.sh
for finetuning Flan-T5-Large
on our data
- Please refer to
analysis/
for all our code regarding analysis of our predicted frames analysis/analysis_data/
contains accompanying preprocessed files frame analysis, note that extended NER predictions and accompanying file for analysis 4.3 in the paper is hosted on huggingface
@article{ranjit2024oath,
title={OATH-Frames: Characterizing Online Attitudes Towards Homelessness with LLM Assistants},
author={Ranjit, Jaspreet and Joshi, Brihi and Dorn, Rebecca and Petry, Laura and Koumoundouros, Olga and Bottarini, Jayne and Liu, Peichen and Rice, Eric and Swayamdipta, Swabha},
journal={arXiv preprint arXiv:2406.14883},
year={2024}
}