Personalized user engagement modeling for mobile videos

Abstract

The ever-increasing mobile video services and users’ demand for better video quality have boosted research into the video Quality-of-Experience. Recently, the concept of Quality-of-Experience has evolved to Quality-of-Engagement, a more actionable metric to evaluate users’ engagement to the video services and directly relate to the service providers’ revenue model. Existing works on user engagement mostly adopt uniform models to quantify the engagement level of all users, overlooking the essential distinction of individual users. In this paper, we first conduct a large-scale measurement study on a real-world data set to demonstrate the dramatic discrepancy in user engagement, which implies that a uniform model is not expressive enough to characterize the distinctive engagement pattern of each user. To address this problem, we propose PE, a personalized user engagement model for mobile videos, which, for the first time, addresses the user diversity in the engagement modeling. Evaluation results on a real-world data set show that our system significantly outperforms the uniform engagement models, with a 19.14% performance gain.