Prompt Engineering for Spanish Sexism Detection

Sexist content has become increasingly prevalent across the internet, including on platforms like X (formerly Twitter), which has 436 million monthly active users. Sexism is defined as discrimination or prejudice based on gender and disproportionately affects women. To combat this, using artificial intelligence (AI) to automatically identify and address sexism in tweets has been proposed as a way to combat this form of violence against women. With the rise of large language models (LLMs), prompt engineering has become crucial to take advantage of their capabilities. It involves designing prompts, that is, specific instructions, to guide models toward desired outputs. Effective prompts significantly impact response quality. In this study, we investigate the impact of using two LLMs with Spanish language support and different prompt formulations on the performance of sexism detection in tweets. Our study used the EXIST 2024 dataset, which contains a variety of sexist and non-sexist tweets in both English and Spanish. We focused our analysis on the 1159 Spanish-language tweets classified as either Sexist and Non-sexist where all annotators were in agreement.