Exploring the impacts of Danmu on online video consumption behaviour: A data-driven method

With the rising popularity of Danmu comments, related system operations have become increasingly crucial for online video-sharing platforms. For the first time, we compare two online video consumption behaviours based on Danmu comments. To achieve this, we quantitatively explored the relationship between Danmu comments and platform users’ viewing/awarding behaviour using real-world data from bilibili.com, a popular Chinese video-sharing platform. Subsequently, we applied grid search, a machine learning method, to detect the threshold of Danmu comments, which represented a turning point in this relationship. According to the comparison of performance-scoring metrics for various models, the piecewise regression rather than quadratic multivariate regression was selected for our estimation. We found that there is an inverted U-shaped relationship along with the threshold frequency of Danmu comments, which triggered changes in the comments’ overall impact. This threshold was lower for awarding behaviour than for viewing behaviour: users who gave video awards devoted greater effort and were thus more critical of Danmu comments. These results have implications for online video-sharing platforms designing and adopting the Danmu system.