FineMuSe: a Fine-Grained Multimodal Dataset for Sexism Detection in Social Media Videos

Description Online sexism manifests in multifaceted forms, including the promotion of toxic masculine behaviors, the denial of discrimination against women, and the reinforcement of body standards to which women are expected to conform. Extensive research has examined sexism in textual content, using data collected from platforms such as X and Reddit. More recently, sexism has begun to be investigated from a multimodal perspective, particularly through studies of memes that analyze the interplay between textual and visual information. However, sexism in videos, which requires considering the interaction among textual, acoustic, and visual information, remains an emerging area of research, mainly due to the limited availability of high-quality datasets. To address this gap, we introduce FineMuSe, the first multimodal Spanish dataset to provide both binary and fine-grained annotations of sexism, users’ responses to sexist content, and the use of rhetorical devices such as irony and humor in both sexist and non-sexist content. FineMuSe is a multisource dataset comprising 828 videos collected from different social media platforms. Its multisource design enables cross-platform comparisons of how sexism is expressed online and how users respond to sexist content.