Understanding Danmaku and Comment Interactions Through Content Features and Video Popularity

This study explores how user interaction with content relates to video popularity on Bilibili, a video-sharing platform. We focus on two primary forms of interaction, danmaku and traditional comment sections, to examine how different content-level characteristics may influence user engagement outcomes. Using an exploratory content analysis framework, we assess six dimensions of potentially controversial expression, quantified through the Perspective API and cross-validated with ChatGPT-4o to address language limitations. A composite popularity index is constructed by integrating views, likes, favorites, danmaku density, and comment density. Hierarchical regression results suggest that interaction volume is the strongest predictor of video popularity, while content-related signals contribute additional explanatory power. Specifically, the fluctuation of overall toxicity in danmaku is positively associated with video popularity, whereas fluctuations in identity attacks show a negative association. In traditional comments, higher average levels of identity attacks and insults are significantly negatively correlated with video popularity. This study contributes by: (1) introducing a multidimensional exploratory framework to assess content-level expression in user interactions; (2) identifying distinct patterns between danmaku and comments in their associations with popularity; and (3) offering implications for content moderation and interaction design in video-sharing environments.