This research introduces MuSeD, a novel multimodal dataset that enables more effective detection of sexist content in Spanish social media videos.

• Combines text, audio, and visual data to improve sexism detection accuracy
• Develops specialized annotation protocols and definitions for multimodal sexism classification
• Establishes benchmarks for future research in multilingual, multimodal content moderation
• Addresses critical security gaps in detecting harmful content across different communication channels

This work has significant implications for online safety systems, providing platforms with more sophisticated tools to identify and mitigate discriminatory content that traditional text-only approaches might miss.

MuSeD: A Multimodal Spanish Dataset for Sexism Detection in Social Media Videos