<p>With the rapid development of network technologies, platforms featuring on-screen, real-time interactive comments—known as danmaku (i.e., comments that scroll across the screen while the video plays)—such as Bilibili, have become important channels for knowledge acquisition. Their highly interactive nature provides a useful context for examining learners’ behavioral and psychological mechanisms in informal learning. This study aims to identify the key determinants of users’ continuance learning intention through a two-stage research design. In Stage One, grounded in established information systems theories, large-scale danmaku and comment texts were analyzed using the Latent Dirichlet Allocation (LDA) and Biterm Topic Model (BTM) to extract core themes and contextual psychological factors, which were subsequently incorporated into an extended structural model. In Stage Two, a questionnaire survey and structural equation modeling were employed to validate the proposed model and examine the pathways among latent variables. The results indicate that social presence does not directly predict continuance learning intention but indirectly influences it through flow experience and user satisfaction. Perceived interactivity and content quality positively affect continuance learning intention directly or indirectly, while emotional value significantly moderates the pathway from perceived interactivity to continuance learning intention. The findings provide new empirical evidence for understanding psychological mechanisms in interactive online learning environments and offer practical implications for optimizing content design and user engagement on knowledge-based video platforms.</p>