감성분석과 유사도를 이용한 한국어 혐오 표현 분류 시스템

Abstract

This paper studies the impact of emotions and similarity on hate speech classification in social networks, especially in personal reviews. Various deep learning models for text sequences have demonstrated notable success in text classification tasks. However, as the trend of writing personalized reviews and social network posts grows, shifts in language usage have emerged. Hate speech in online reviews has become a significant social issue, leading individuals to experience anxiety, loneliness, depression, or even suicidal thoughts. In this study, we aim to integrate sentiment analysis into hate speech classification to examine its effects on detecting toxic language in text reviews. To achieve this, we utilized a dataset that includes multi-class emotion classification. In addition, we incorporated similarity analysis for individual sentences, applying it through our proposed weighted method. For hate speech recognition, we conducted experiments using an open Korean dataset. Our results indicate that incorporating our proposed method into hate speech recognition slightly improved the F1-score. Furthermore, we analyzed the correlation between hate speech and emotions through topic modeling.