电信科学 ›› 2021, Vol. 37 ›› Issue (8): 18-26.doi: 10.11959/j.issn.1000-0801.2021116

• 研究与开发 • 上一篇    下一篇

基于神经网络与马尔可夫组合模型的视频流行度预测算法

马学森1, 陈树友1, 许向东2, 储昭坤1   

  1. 1 合肥工业大学计算机与信息学院，安徽 合肥 230601
    2 中国电信股份有限公司研究院，广东 广州 510630
  • 修回日期:2021-06-12 出版日期:2021-08-20 发布日期:2021-08-01
  • 作者简介:马学森（1976− ），男，合肥工业大学计算机与信息学院副教授，主要研究方向为移动边缘计算、无线网络
    陈树友（1995− ），男，合肥工业大学计算机与信息学院硕士生，主要研究方向为高可靠性分布式系统、移动边缘计算
    许向东（1978− ），男，现就职于中国电信股份有限公司研究院，主要研究方向为移动通信、5G智慧应用、物联网应用、大数据及AI系统研发、网络运营技术等
    储昭坤（1996− ），男，合肥工业大学计算机与信息学院硕士生，主要研究方向为移动边缘计算、无线网络
  • 基金资助:
    国家重点研发计划项目(2020YFC1512601);广东省科技发展专项基金资助项目(2017A010101001);安徽省教育厅高等学校省级质量工程(2019MOOC020);中央高校基本科研业务费专项基金资助项目(PA2019GDKPK0079)

Neural network and Markov based combination prediction algorithm of video popularity

Xuesen MA1, Shuyou CHEN1, Xiangdong XU2, Zhaokun CHU1   

  1. 1 School of Computer and Information, Hefei University of Technology, Hefei 230601, China
    2 Research Institute of China Telecom Co., Ltd., Guangzhou 510630, China
  • Revised:2021-06-12 Online:2021-08-20 Published:2021-08-01
  • Supported by:
    The National Key Research ＆ Development Program of China(2020YFC1512601);Science and Technology Development Special Foundation of Guangdong Province(2017A010101001);Colleges ＆ Universities Quality Engineering Project of Education Department of Anhui Province(2019MOOC020);Fundamental Research Funds for the Central Universities(PA2019GDKPK0079)

在线阅读

34

PDF下载

233

摘要：

为了提升用户体验，降低运营商的成本，将播放最多的视频内容提前放入用户侧缓存是业界的通用做法，如何有效预测视频播放热度已经成为业界热点问题。针对传统预测算法非线性映射能力差、预测精度低及自适应性弱等缺点，提出基于神经网络与马尔可夫组合模型的视频流行度预测算法（Mar-BiLSTM），该算法通过构建双向长短期记忆（bi-directional long short-term memory，BiLSTM）网络模型可以保留时间序列两个方向的信息依赖；同时在避免引入外部变量导致模型复杂度增加的情况下，利用马尔可夫性质进一步提高了模型的预测精度。实验结果表明，与传统的时间序列和经典的神经网络算法相比，所提算法提升了视频流行度预测的准确性、时效性，并降低了计算量。

关键词: 流行度预测, 内容缓存, 双向长短期记忆网络, 马尔可夫

Abstract:

Caching popular video into user-side in advance improves the user experience and reduces operator costs, which is a common practice in the industry.How to effectively predict the popularity of videos has become a hot issue in the industry.On account of the shortcomings of traditional prediction algorithms such as poor nonlinear mapping ability, low prediction accuracy and weak adaptability, a video popularity prediction algorithm based on a neural network and Markov combined model (Mar-BiLSTM) was proposed.Information dependencies were preserved by constructing bidirectional memory network model (bi-directional long short-term memory, BiLSTM), the prediction accuracy of the model was further improved by using Markov properties while avoiding the increase of the complexity of the model caused by the introduction of external variables.Experimental results show that compared with traditional time series and classic neural network algorithms, the proposed algorithm improves predicting accuracy, effectiveness and reduces the amount of calculation.

Key words: popularity prediction, content caching, BiLSTM network, Markov

中图分类号: 

No Suggested Reading articles found!