ResponSight: Explainable and Collaborative Moderation Approach for Responsible Video Content in UGC Platform
Haonan Nan, Zixiao Wang, Yu Zhang, Xuantao Zhang, Saixing Zeng · 2025
The rapid spread of user-generated video content on social media has raised pressing ethical concerns, requiring a holistic approach to moderate the contents. While current research primarily focuses on explicit text-content violations detection, it overlooks the ethical dimensions of content responsibility and users’ needs for clear explanations and specific guidance. In response, we construct a novel video moderation system for user-generated content (UGC) platforms, shifting from traditional reactive approaches to a proactive, distributed human-machine workflow. We design a detection tool framework called ResponSight that collaborates with users to promote responsible content creation. ResponSight contains two core modules: the explainable evaluation module and the adaptive suggestion module, which driven by multimodal large language models (MLLMs). By offering transparent, explainable feedback, this system enables users to understand the rationale behind moderation decisions and refine their videos to align with ethical and social responsibility standards.