Chain-of-Thought-Embeddings

This repository contains code for the 2023 EMNLP Findings paper Chain-of-Thought Embeddings for Stance Detection on Social Media

In the data folder, we provide the data for Tweet-Stance along with the ChatGPT (gpt-3.5-turbo) COT outputs used in this work. The details of our prompting strategy can be found in the paper.

How to Run Code

First, install requirements

pip install -r requirements.txt

Then, you can run code as follows:

python main.py <modality> --seed <seed> 

Valid args for include text-only, cot-only, text+cot. We used seeds = [1,2,3,4,5] in Table 7 of our paper.

Note

We are unable to release data for presidential-stance, but you can sign a terms of use agreement and download it here. If you are interested in details re: our work with this dataset please reach out! joseph {dot} m {dot} gatto {dot} gr @ dartmouth {dot} edu