A Video-Quality Driven Strategy in Short Video Streaming

Abstract

In recent years, fueled by the rapid advances in high-speed mobile networks, streaming short-form videos over mobile devices (e.g., TikTok) is ubiquitous among mobile users. Despite its widespread application, our investigation revealed that the video quality experienced by the viewers exhibits substantial fluctuations over different short video sessions, so the Quality-of-Experience (QoE) is in fact far from optimal. To tackle this problem, this work develops a novel strategy called Quality Aware Short Video Streaming (QASVS), which employs deep reinforcement learning to automatically generate a quality-driven bitrate decision model to dynamically determine the bitrate for each session. Extensive evaluations show that QASVS can maintain much more consistent video quality across the video sessions. Moreover, compared to the state-of-the-art algorithms, it can improve the video quality by 11.1%∼27.9% and significantly reduce the rebuffering events. Therefore, QASVS offers an effective solution for high-performance short video services.