This research introduces a novel approach to moderate real-time comments (Danmaku) on video platforms by combining impact captions with large language models to reduce toxic behavior.

• Creates a proactive moderation system that can identify and prevent inappropriate comments
• Leverages visual techniques to improve user engagement while reducing harmful content
• Enhances social interactions on video platforms through innovative security measures
• Addresses the growing challenge of toxic behavior in real-time comment systems

From a security perspective, this approach offers video platforms a way to maintain user engagement features while proactively addressing content moderation challenges—crucial for creating safer online communities without sacrificing social interaction.

DanModCap: Designing a Danmaku Moderation Tool for Video-Sharing Platforms that Leverages Impact Captions with Large Language Models