Predicting the Popularity of DanMu-enabled Videos: A Multi-factor View

Recent years have witnessed the prosperity of a new type of real-time user-generated comment, or so-called DanMu, in many recent online video platforms. These DanMu-enabled video platforms present scrolling marquee comments overlaid directly on top of the videos by synchronizing these comments to specific playback times. In this paper, we study the prediction of video popularity in these platforms, which may benefit a lot of applications ranging from online advertising for website holders to popular video recommendation for audiences. Different from traditional online video platforms where only traditional reviews are available, these DanMus make viewers easily see other viewers’ opinions and communicate with each other in a much more direct way. Consequently, viewers are easily influenced by others’ behaviors over time, which is considered as the herding effect in social science. However, how to address the unique characteristics (i.e., the herding effect) of DanMu-enabled online videos for more accurate popularity prediction is still under-explored. To that end, in this paper, we first explore and measure the herding effect of DanMu-enabled video popularity from multiple aspects, including the popular videos, the popular DanMus and the newly updated videos. Also, we recognize that the uploaders’ influence and video quality affect the video popularity as well. Along this line, we propose a model that incorporates the herding effect, uploaders’ influence and video quality for predicting the video popularity. An effective estimation method is also proposed. Finally, experimental results on real-world data show that our proposed prediction model improves the prediction accuracy by 47.19 % compared to the baselines.