Korean Hate Speech Detection Using Hate-Rationale Word Prediction

This paper presents a novel approach for Korean hate speech detection that enhances model performance by incorporating hate-rationale word prediction as an auxiliary task. Our method identifies and leverages specific words that contribute to hateful expressions.

The proposed framework combines hate speech detection with rationale word prediction to improve both performance and interpretability. By learning to identify words that contribute to hate speech, the model achieves better detection accuracy while providing insights into the reasoning behind its decisions. Experimental results on Korean social media datasets demonstrate significant improvements over baseline approaches.

Recommended citation: Yunseo Choi, Byounghan Lee, Kyung-Ah Sohn. (2023). "Korean Hate Speech Detection Using Hate-Rationale Word Prediction." Korea Software Congress (KSC2023), Busan, South Korea.