# Outcome-Constrained Large Language Models for Countering Hate Speech

Source: [Outcome-Constrained Large Language Models for Countering Hate Speech](<https://www.sota2.com/research/paper/68fa75061e209131cea6555f>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-26T03:24:17.272Z

Authors: Lingzi Hong, Pengcheng Luo, Eduardo Blanco, Xiaoying Song

Published: 2024-03-25T19:44:06.000Z

## Abstract

Automatic counterspeech generation methods have been developed to assist efforts in combating hate speech. Existing research focuses on generating counterspeech with linguistic attributes such as being polite, informative, and intent-driven. However, the real impact of counterspeech in online environments is seldom considered. This study aims to develop methods for generating counterspeech constrained by conversation outcomes and evaluate their effectiveness. We experiment with large language models (LLMs) to incorporate into the text generation process two desired conversation outcomes: low conversation incivility and non-hateful hater reentry. Specifically, we experiment with instruction prompts, LLM finetuning, and LLM reinforcement learning (RL). Evaluation results show that our methods effectively steer the generation of counterspeech toward the desired outcomes. Our analyses, however, show that there are differences in the quality and style depending on the model.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Crisis response generation on Earthquake and Typhoon cross-crisis scenarios](<https://www.sota2.com/research/sota/crisis-response-generation-on-earthquake-and-typhoon-cross-crisis-scenarios.md>): Crisis response generation; Earthquake and Typhoon cross-crisis scenarios generalization Llama-3.1-8B-Instruct; Professionalism Score: 0.97.

- [Crisis Response Generation on Twitter](<https://www.sota2.com/research/sota/crisis-response-generation-on-twitter.md>): Crisis response generation; Twitter; Professionalism: 75.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2403.17146>)

[PDF](<https://arxiv.org/pdf/2403.17146.pdf>)