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

TL;DR

DanModCap addresses the challenge of moderating Danmaku on live video-sharing platforms by introducing Impact Captions—generative, resonance-driven captions that guide viewer interpretation and behavior. The approach combines a taxonomy of caption styles with a three-stage generation pipeline using $LDA$, $BERT$, $LLaMA2$, and $Stable Diffusion$ to produce text and visuals that foster cognitive and emotional engagement while steering conversations toward prosocial Danmaku. An expert co-design workshop informs design goals and a web-based prototype demonstrates real-time generation and display of captions, followed by a lab-based user study showing reduced negative interactions and enhanced community cohesion. The work highlights both the potential and the risks of AI-driven proactive moderation in large-scale, live-context settings, emphasizing the need for culturally sensitive, interpretable interventions and rigorous evaluation before deployment.

Abstract

Online video platforms have gained increased popularity due to their ability to support information consumption and sharing and the diverse social interactions they afford. Danmaku, a real-time commentary feature that overlays user comments on a video, has been found to improve user engagement, however, the use of Danmaku can lead to toxic behaviors and inappropriate comments. To address these issues, we propose a proactive moderation approach inspired by Impact Captions, a visual technique used in East Asian variety shows. Impact Captions combine textual content and visual elements to construct emotional and cognitive resonance. Within the context of this work, Impact Captions were used to guide viewers towards positive Danmaku-related activities and elicit more pro-social behaviors. Leveraging Impact Captions, we developed DanModCap, an moderation tool that collected and analyzed Danmaku and used it as input to large generative language models to produce Impact Captions. Our evaluation of DanModCap demonstrated that Impact Captions reduced negative antagonistic emotions, increased users' desire to share positive content, and elicited self-control in Danmaku social action to fostering proactive community maintenance behaviors. Our approach highlights the benefits of using LLM-supported content moderation methods for proactive moderation in a large-scale live content contexts.

Figures (3)

• Figure 1: The Taxonomy of Impact Captions that was identified from our video-based analysis of popular online variety shows.
• Figure 2: DanModCap's Impact Caption Generation Model, which used LDA to extract topics from Danmaku and supervised learning on a pre-trained model to identify Danmaku sentiment. Next, LLaMA2, which incorporates prompt engineering, was used to generate moderation text. Following this, Stable Diffusion generated a speech bubble.
• Figure 3: DanModCap Interface including the Admin Control, an example Impact Caption, the Video Content, and the Danmaku Editor.