Reproduce “When Rules Fall Short: Agent-Driven Discovery of Emerging Content Issues in 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
Chenghui Yu, Hongwei Wang, Junwen Chen, Zixuan Wang, Bingfeng Deng, Zhuolin Hao, Hongyu Xiong, Yang Song
Researchers from TikTok Inc. developed an MLLM-based agent to automate the discovery of emerging content issues on short-video platforms, accelerating policy updates and enhancing content governance. The system reduced the time for policy iteration by 70.2% and improved the effectiveness of governance models by decreasing problematic video views by 14.9% in online A/B tests.
Key insights
- MLLM agents can effectively leverage extensive world knowledge and multimodal understanding to perform complex, multi-stage tasks in content moderation without extensive prior labeled data.
- Integrating auxiliary tools like lightweight governance models and multimodal embeddings significantly boosts the agent's accuracy in distinguishing covered issues from emerging variants or entirely new issues.
- By focusing on the "essential decision logic" of content policies rather than strict literal definitions, the agent can proactively identify new forms of problematic content that fall outside current explicit rules.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.