LLM in the Loop: Creating the ParaDeHate Dataset for Hate Speech Detoxification

TL;DR

This paper introduces ParaDeHate, a large-scale hate speech detoxification dataset created using an LLM-in-the-loop pipeline with GPT-4o-mini, enabling scalable, high-quality data generation for detoxification tasks.

Contribution

It presents a novel LLM-based pipeline for creating detoxification datasets and releases ParaDeHate, the first large-scale hate speech detoxification benchmark.

Findings

LLM performs comparably to human annotators in detoxification tasks.

Fine-tuning models like BART on ParaDeHate improves detoxification performance.

ParaDeHate enables scalable and effective hate speech detoxification evaluation.

Abstract

Detoxification, the task of rewriting harmful language into non-toxic text, has become increasingly important amid the growing prevalence of toxic content online. However, high-quality parallel datasets for detoxification, especially for hate speech, remain scarce due to the cost and sensitivity of human annotation. In this paper, we propose a novel LLM-in-the-loop pipeline leveraging GPT-4o-mini for automated detoxification. We first replicate the ParaDetox pipeline by replacing human annotators with an LLM and show that the LLM performs comparably to human annotation. Building on this, we construct ParaDeHate, a large-scale parallel dataset specifically for hatespeech detoxification. We release ParaDeHate as a benchmark of over 8K hate/non-hate text pairs and evaluate a wide range of baseline methods. Experimental results show that models such as BART, fine-tuned on ParaDeHate, achieve better performance in style accuracy, content preservation, and fluency, demonstrating the effectiveness of LLM-generated detoxification text as a scalable alternative to human annotation.