基于长短期记忆的流媒体QoE预测模型  

Modeling QoE prediction of media streaming based on long and short term memory

作　　者：孙鹏 白光伟 沈航 顾一鸣 SUN Peng;BAI Guang-wei;SHEN Hang;GU Yi-ming(College of Computer Science and Technology,Nanjing Tech University,Nanjing 211816,China)

机构地区：[1]南京工业大学计算机科学与技术学院,江苏南京211816

出　　处：《计算机工程与设计》2020年第6期1501-1508,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61073197);南京市科技计划基金项目(201608009)。

摘　　要：现阶段关于连续性QoE预测模型研究存在着实时性差和精确度不高的问题。针对这些问题,提出L-QoE模型,一种使用长短期记忆(LSTM)网络的基于递归神经网络的QoE预测模型,用于捕获时变QoE中涉及的非线性和复杂时间依赖性。基于公开可用的连续QoE数据库的评估,实验结果表明,L-QoE模型能够有效模拟QoE动态性,得出更精确的预测结果。将所提模型与最近提出的QoE预测模型进行比较,通过对比得出该模型在数据库中突出的性能。At present,there are problems of poor real-time performance and low accuracy in the study of continuous QoE prediction models.To solve these problems,an L-QoE model,a recursive neural network-based QoE prediction model using long-term and short-term memory(LSTM)networks,was used to capture the nonlinear and complex time dependence involved in time-varying QoE.Based on the evaluation of publicly available continuous QoE databases,the experimental results reflect that the L-QoE model can effectively simulate QoE dynamics and produce more accurate prediction results.And this model compared with the recently proposed QoE prediction model shows its outstanding performance in the database.

关 键 词：重载 比特率波动 LSTM神经网络 主观体验质量(QoE) 时间序列预测 非马尔可夫动态性建模 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]