KoSBi: A Dataset for Mitigating Social Bias Risks Towards Safer Large Language Model Application

TL;DR

This paper introduces KoSBi, a Korean social bias dataset with 34,000 pairs covering 72 demographic groups, and demonstrates that filtering moderation can significantly reduce biases in large language models.

Contribution

The paper provides a new localized social bias dataset for Korean LLMs and shows effective bias mitigation through filtering-based moderation techniques.

Findings

Bias reduction of 16.47%p in HyperCLOVA models

KoSBi covers 72 demographic groups in 15 categories

Filtering moderation effectively reduces social biases

Abstract

Large language models (LLMs) learn not only natural text generation abilities but also social biases against different demographic groups from real-world data. This poses a critical risk when deploying LLM-based applications. Existing research and resources are not readily applicable in South Korea due to the differences in language and culture, both of which significantly affect the biases and targeted demographic groups. This limitation requires localized social bias datasets to ensure the safe and effective deployment of LLMs. To this end, we present KO SB I, a new social bias dataset of 34k pairs of contexts and sentences in Korean covering 72 demographic groups in 15 categories. We find that through filtering-based moderation, social biases in generated content can be reduced by 16.47%p on average for HyperCLOVA (30B and 82B), and GPT-3.