Reproduce “Complementing Text Based Moderation with Multimodal Grounding: Detecting Fraudulent Inducement Behaviors on Short Video Platforms” 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
Xiaoyong Yu
The authors develop a video moderation system for internet-finance inducements that predicts event categories and approximate time intervals. It retains OCR transcripts while adding a multimodal model to detect visual cues that text-only moderation may miss. On a rebalanced test set, SFT+RL reaches 67.0% macro recall and 60.1% non-textual recall, while RL-only has stronger localization; neither leads on both.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.