Characterizing and Predicting the Popularity of Online Videos

Abstract

A huge amount of video content has been generated on the Internet, and user attention among those videos is allocated in an asymmetric way, with the vast majority barely noticed while a few of videos become very popular. Hence, understanding the popularity characteristics of online videos and predicting the future popularity of individual videos are of great importance. They have direct implications in various contexts, such as service design, advertisement planning, network management, and so on. In this paper, we address those two problems head-on based on data collected from a leading online video service provider in China, namely Youku. We firstly analyze the characteristics of Youku video popularity from four key aspects: long-term popularity, video lifetime, popularity evolution pattern, and early stage popularity. Then, we undertake the challenge of future popularity prediction, by proposing a model that can capture the popularity dynamics based on early popularity evolution pattern and future popularity burst prediction. The approach is validated on exhaustive real-world data and achieves significant decreases in relative prediction errors, reaching up to 32.73% and 11.28% reductions over two state-of-the-art baseline models, respectively. At last, we also provide the potential and limitation analysis of model parameters in practice.