A Statistical-based Rate Adaptation Approach for Short Video Service

Dynamic adaptive streaming has been recently widely adopted for providing uninterrupted video streaming services to users with dynamic network conditions and heterogeneous devices in Live and VoD (Video on Demand). However, to the best of our knowledge, no rate adaptation work has been done for the new arisen short video service, where a user generally watches many independent short videos with different contents, quality, bitrate, and length (generally about several seconds). In this work, we are the first to study the rate adaptation problem for this scenario and a Statistical-based Rate Adaptation Approach (SR2A) is proposed. In SR2A, each short video is transcoded into several versions with different bitrate. Then, when a user watches the short videos, the network conditions and player status are collected, and together with the to be requested video’s information, the best video version (bitrate or quality) will be selected and requested. Thus, the user will experience the short videos with the most suitable quality depending on the current network conditions. We have collected the network trace and user behavior data from Kuaishou1, the largest short video community in China. By the collected data set, the users’ watching behavior is analyzed, and a statistical model is designed for bandwidth prediction. Then, combined with the video information derived from the manifest, the maximal video bitrate is selected under the condition that the probability of play interruption is smaller than a predefined threshold during the whole playback process. The trace based experiments show that SR2A can greatly improve the user experience in quality and fluency of watching short videos.