Modeling and quantifying the forces driving online video popularity evolution

Abstract

Video popularity is an essential reference for optimizing resource allocation and video recommendation in online video services. However, there is still no convincing model that can accurately depict a video's popularity evolution. In this paper, we propose a dynamic popularity model by modeling the video information diffusion process driven by various forms of recommendation. Through fitting the model with real traces collected from a practical system, we can quantify the strengths of the recommendation forces. Such quantification can lead to characterizing video popularity patterns, user behaviors and recommendation strategies, which is illustrated by a case study of TV episodes.