Deep Learning-based Short Video Recommendation and Prefetching for Mobile Commuting Users

Abstract

Mobile short video application is growing rapidly and it is quickly occupying people's life. In this paper, we consider an emerging yet common scenario of short video application usage: mobile users watching short videos on their daily commuting trip on high speed public transport, where the network condition is unsatisfactory. To reduce users waiting time and improve the QoE, we propose a deep learning-based data recommendation and prefetching scheme which obtains user interests and pushes the preferred short video content to the most likely base station that users will be connected to. We use Principal Component Analysis (PCA) plus dropout to reduce the feature dimensions of Inception structure to improve the short video recommendation speed without degrading the accuracy. Through experimental evaluations, we show that the proposed scheme can effectively recommend short video and predict user trajectory, with a recall rate of 100%.