PoliticalBias Reposiotry

This repository contains my personal module report for DCU, 2024. The experiments are reproduced from From Pretraining Data to Language Models to Downstream Tasks: Tracking the Trails of Political Biases Leading to Unfair NLP Models presented at ACL 2023. I utilized newer models, Llama-3 and GPT-4o, for comparison, and fine-tuned GPT-2 using the POLITICS news dataset.

Experiment of Stage 1

Code of this part can be found in PoliLean Public. And the requirement of enviroment as well.

Datasets used to fine-tune

For datasets used to fineture, visit POLITICS.

fine_tune_code

This folder contains two Python scripts: one for generating a smaller training set and another for fine-tuning the model.

PCT

Prompts, statements, example response and score are provided in this folder.