Researchers have developed a system for identifying and characterizing sexism in multimodal content like memes and short-form videos. Their approach combines visual, textual, and LLM-derived semantic features, feeding them into gradient-boosted regression models. The study found that LLM-derived cues significantly improved sexism identification in memes, while video analysis proved sensitive to feature selection and cross-modal noise, suggesting a need for more robust temporal modeling in video content. AI
IMPACT This research demonstrates the utility of LLMs in identifying nuanced harmful content, potentially improving AI safety tools for content moderation.
RANK_REASON The cluster contains a research paper detailing a new methodology for multimodal sexism identification.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
