Researchers have developed CLARA, a novel framework designed to detect hateful content in videos by analyzing them at the clip level. This approach models videos as sequences of fine-grained clips to better capture temporally localized hateful signals, which are often brief and implicit. CLARA incorporates a Mixture-of-Experts clip encoder for multimodal alignment, a contrastive objective for modeling short-term and long-range temporal dependencies, and VLM-derived rationales to provide semantic guidance. Experiments on three datasets show CLARA significantly outperforms existing methods. AI
IMPACT This framework could improve the safety and moderation capabilities of video-centric social media platforms.
RANK_REASON This is a research paper detailing a new framework for hateful video detection. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- CatalyzeX
- CLARA
- Connected Papers
- DagsHub
- Gotit.pub
- Hugging Face
- Litmaps
- mixture of experts
- ScienceCast
- Scite
- Transformer++
- vision-language model
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
