From Watching to Wishing: A Multimodal Computational Analysis of How PUGC Video-Danmaku Ecology Shapes Travel Intention

Digital tourism marketing increasingly relies on platform user-generated content (PUGC), yet mechanisms through which multimodal videos and real-time social interactions shape travel decisions remain largely opaque. This study investigates this "black box" by analyzing 2,650 tourism videos from Bilibili and their synchronized Danmaku (bullet-screen comments) through an innovative computational framework integrating computer vision, natural language processing, and machine learning. Three significant findings emerge: First, cognitive features demonstrate exceptional predictive power for travel intention (42.2% of model importance from only 10.4% of features), while traditional audiovisual features show limited contribution (16.1% importance from 45.8% of features)—questioning visual-centric marketing assumptions. Second, non-linear models achieve substantially higher explanatory power than linear approaches (R² = 0.518 vs. 0.033), revealing threshold effects and cognitive gating mechanisms not captured by additive frameworks. Third, Danmaku appears to function as cognitive scaffolding rather than distraction, with Granger causality suggesting orchestrated attention cascades from visual to audio to textual processing (p < 0.001). By operationalizing previously unobservable mechanisms through extraction of 100 + multimodal features aligned with 475 million Danmaku comments, this study provides new insights into platform affordances and tourism decision-making. The findings suggest important implications for digital tourism strategies: consider prioritizing cognitive activation alongside production aesthetics, facilitating synchronized social interaction, and recognizing that mental engagement may constitute a critical resource in attention economies.