Reproduce “Dynamic Content Moderation in Livestreams: Combining Supervised Classification with MLLM-Boosted Similarity Matching” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Wei Chee Yew, Hailun Xu, Sanjay Saha, Xiaotian Fan, Hiok Hian Ong, David Yuchen Wang, Kanchan Sarkar, Zhenheng Yang and 1 more

The system moderates livestreams by splitting them into 20-second clips and routing them to a preset-violation classifier and a retrieval path that compares clips with historical violations. A large multimodal model teaches smaller models during training, rather than running online. In long-term backtests, classification reduced unwanted views about 4%, and retrieval reduced them a further 2–4%; this does not establish detection of every novel violation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.