MilaNLProc/weigh-your-own-words

Code associated with the paper "Weigh Your Own Words: Improving Hate Speech Counter Narrative Generation via Attention Regularization"

Weigh Your Own Words: Improving Hate Speech Counter Narrative Generation via Attention Regularization

This repository contains the code to replicate our experiments.

Getting Started

Create a new Python environment and install the required packages with

pip install -r requirements

Fine-Tuning

Use finetune_gpt2.py to fine-tune a GPT2 model using KLAR or EAR.

Attention Regularization

Please refer to the src folder to find our custom GPT2 implementation and code to compute KLAR and EAR attention-based losses.

Reference

TBD