A new popularity prediction model based on lifetime forecast of online videos
Benle Su, Yumei Wang, Yu Liu · 2016
Along with the rapid development of bandwidth and easier access into the Internet, online video traffics have been increasing dramatically and attracting a huge number of customers. Predicting the video popularity is of great importance for various applications, such as precise advertisement delivery, reasonable cache allocation, video recommendation and etc. In this paper, we observe the relationship between video lifetime and video popularity and find that longer lifetime usually means stronger future popularity. So we introduce video lifetime as a coefficient in the popularity prediction model and propose a multi-linear model based on the historical view count, future burst state and video lifetime to predict future video popularity. Our model is validated with the data extracted from Youku, a popular video service provider in China, and it is proved to outperform a state-of-the-art baseline model with a 1.58% reduction of the relative prediction errors.