网络视频服务中用户体验质量预测研究

发布时间：2018-02-10 01:51

本文关键词： 测量网络视频服务用户体验质量机器学习　出处：《北京交通大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着互联网技术以及视频多媒体技术的不断发展,网络视频作为一种重要的休闲娱乐方式,受到了人们的一致追捧。思科公布的互联网预测报告显示:2015年网络视频流量占全部互联网流量的70%,预计到2020年所有消费的网络流量中的视频流量将占到82%,其中移动视频数据流量将占总网络流量的50%。如此庞大的视频数据流量对当前的视频服务,特别是移动端视频服务,带来了极大的挑战。与此同时,视频用户对视频观看质量也提出了更高层次的要求:高视频分辨率、低启动时延、低缓冲率,追求更高的用户体验质量(Quality of Experience,QoE)。因此,研究如何精准预测网络视频服务中的用户体验质量,近而提升视频用户体验质量,具有很大的理论价值和商业应用价值。现有的关于用户体验质量的研究工作中,大多是研究视频用户观看行为以及视频质量影响因素,或者提出一些复杂的控制平台系统来优化网络视频资源传输效率,或者研究复杂的视频编码,来提升用户体验质量。本文拟运用机器学习算法,构建简单、易部署的基于用户终端的QoE模型,提升用户体验质量。本文的具体贡献主要有如下四个方面。(1)详细分析了 PPTV视频用户接入日志数据集,发现:1)起始缓冲时长比缓冲总时长更需要针对性的优化;2)缓冲次数与用户有效观看时间比的相关性最大。在此基础上设计了一种高性能的基于随机森林算法的QoE映射模型,在预测用户体验质量不好时的F1值达到0.77,并且起始缓冲时长和缓冲次数对模型预测效果的影响较大。(2)开发了一整套适用于LTE网络环境下DASH视频质量研究的实验平台。具体说来,在阿里云服务器上搭建了 DASH视频服务器,并部署了 MongoDB数据库用于测量数据的持久化存储;开发了 Androidapp应用用于采集LTE网络质量参数,修改dashjs客户端源码来采集DASH视频客户端播放信息。(3)通过对实验测量数据的研究分析发现:1)当缓冲区长度低于0.5秒钟时,视频将会出现卡顿;2)当前LTE网络下的DASH视频用户体验质量的主要问题在于往返时间(RTT)。(4)提出了一种基于"时间窗口"的预测方法,设计了两种基于随机森林算法的QoE模型,在预测用户体验质量不好时的F1值达到0.87。并且,最佳的间隔时间窗口值:28秒,最佳的历史时间窗口值为10秒到18秒。
[Abstract]:With the continuous development of Internet technology and video multimedia technology, network video as an important way of leisure and entertainment, In 2015, network video traffic accounted for 70 percent of all Internet traffic, and it is expected that by 2020, video traffic will account for 82 percent of all network traffic consumed. Moving video data traffic will account for 50 percent of the total network traffic. In particular, the mobile video service brings great challenges. At the same time, video users also put forward higher quality requirements for video viewing: high video resolution, low startup delay, low buffering rate. Therefore, research on how to accurately predict the quality of user experience in online video services, and improve the quality of video user experience, It has great theoretical value and commercial application value. Most of the existing research work on the quality of user experience is to study the viewing behavior of video users and the influencing factors of video quality. Or some complex control platform systems are proposed to optimize the transmission efficiency of network video resources, or to study complex video coding to improve the quality of user experience. The QoE model based on user terminal is easy to deploy to improve the quality of user experience. The specific contributions of this paper are as follows: (1) the PPTV video user access log data set is analyzed in detail. It is found that the initial buffer time is more important than the total buffer time to optimize the buffer times. The correlation between the buffer times and the effective viewing time ratio is the greatest. Based on this, a high performance QoE mapping model based on stochastic forest algorithm is designed. The F1 value is 0.77 when the user experience quality is not good, and the effect of the initial buffer time and buffer times on the model prediction effect is great.) A set of experimental platforms suitable for the research of DASH video quality in LTE network environment are developed. DASH video server was built on Ali cloud server, and MongoDB database was deployed for persistent storage of measurement data. Androidapp application was developed to collect LTE network quality parameters. Modify the dashjs client source code to collect the DASH video client playback information.) by analyzing the experimental data, we find that: 1) when the buffer length is less than 0.5 seconds, The main problem of the DASH video user experience quality under the current LTE network is that the round trip time is RTT. 4) A prediction method based on "time window" is proposed, and two QoE models based on stochastic forest algorithm are designed. The F1 value is 0.87 when the user experience quality is not good. Moreover, the optimal interval window value is: 28 seconds, and the best historical time window value is from 10 seconds to 18 seconds.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.09;TN919.8

【参考文献】