Contemporary video repositories increasingly overlay real-time, user-generated metadata in the form of danmaku (bullet comments) onto linear content, creating a complex retrieval environment defined by the tension between collaborative sensemaking and cognitive overload. This study examines this human information interaction dynamic, analyzing how observable creator-related source cues, Content value (CV) signals and peer feedback are associated with engagement within information-dense streams.
Leveraging large-scale behavioral trace data from 261 informational videos and over 130,000 time-anchored comments on Bilibili, the study employs a multi-method approach combining partial least squares structural equation modeling, partial least squares multi-group analysis and BERTopic modeling to triangulate structural engagement associations with semantic discourse patterns.
The analysis identifies association patterns consistent with a sequential cross-route account in which peripheral source cues function as heuristic filters that are linked to CV signals and subsequent feedback interaction. Crucially, this pattern is context-contingent: high-efficacy information environments show stronger content-oriented engagement, whereas low-efficacy contexts show stronger associations with peripheral source heuristics and social signals.
The research refines dual-process theories within digital information management by specifying a measurable, sequential association pattern for observable engagement under information density. It distinguishes Audience Conversion Efficacy as a critical moderator, offering evidence-based strategies for adaptive interface design that align information presentation with content utility.
