Topic Mining and Evolutionary Analysis of Bilibili Learning Video Danmu Based on BERTopic Model

Abstract

In online learning environments, bullet comments (danmu) integrate real-time user discussions with video content, offering valuable data for analyzing learners' states. However, existing analysis methods rely on traditional machine learning frameworks, neglecting semantic features and temporal information. This paper uses the BERTopic model to mine topics from bullet comments in Bilibili learning videos, exploring their distribution, discussion direction, and evolution. Results show that bullet comments aid knowledge construction, emotional support, and reflect learners' needs and emotional shifts. These insights provide valuable guidance for optimizing online learning resources, enhancing teaching design, and improving teaching quality.