1. 서론 2. 성 편향성 데이터 구축 동향 3. KoBiNLI 데이터 설계 및 수집 4. KoBiNLI 데이터 분석 5. 결론
This study proposes a dataset designed to quantitatively assess gender bias. It focuses on linguistic and emotional features within the Korean sociocultural context, incorporating stereotypical occupations and gendered adjectives. To address growing ethical concerns regarding AI, we created a dataset of 1,076 sentences in natural language inference (NLI) format. The analysis of this dataset revealed significant gender biases in vocabulary and emotional expressions between men and women. This study highlighted the importance of culturally relevant datasets in the development of fair and ethical AI systems. |
