• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学

• 人工智能与数据挖掘 • 上一篇    下一篇

基于时间序列的音乐流行趋势预测研究

郁伟生1,邓伟1,张瑶2,李蜀瑜1,2   

  1. (1.陕西师范大学网络信息中心,陕西 西安 710119;2.陕西师范大学计算机科学学院,陕西 西安 710119)
  • 收稿日期:2017-06-14 修回日期:2017-08-15 出版日期:2018-09-25 发布日期:2018-09-25
  • 基金资助:

    国家自然科学基金(41271387);中央高校基本科研业务费项目(GK201703055)

Music popular trends prediction based on time series

YU Weisheng1,DENG Wei1,ZHANG Yao2,LI Shuyu1,2   

  1. (1.Network Information Center,Shaanxi Normal University,Xi’an 710119;
    2.School of Computer Science,Shaanxi Normal University,Xi’an 710119,China)
     

     
  • Received:2017-06-14 Revised:2017-08-15 Online:2018-09-25 Published:2018-09-25

摘要:

在大数据环境下,对音乐及听众的历史数据进行分析,可以实现对音乐流行趋势较为准确的预测。在STL、HoltWinters分解模型的基础上,提出TSMP算法。该算法从长期趋势和周期两方面进行分析,对长期趋势编码和分类并基于类别最优值选择法对音乐流行趋势进行预测。基于TSMP算法,进而提出ETSMP算法,该算法基于子序列模式匹配法及对近期发布新专辑的附加处理,实现更精准的预测。在清华大学和阿里云天池大数据竞赛平台承办的“2016中国高校计算机大赛——大数据挑战赛之阿里音乐流行趋势预测”比赛中,参赛团队凭借提出的ETSMP算法对2016年9月~10月艺人的播放量实现了较好的预测,并在此次比赛中夺得亚军。
 

关键词: 时间序列, 音乐流行趋势, 类别最优值选择, 子序列模式匹配

Abstract:

In big data environment, analyzing the historical data of music and audiences can achieve accurate prediction of music popular trends. Based on the STL and HoltWinters decomposition models, a Time Series based Music Prediction (TSMP) algorithm is proposed. The TSMP algorithm analyzes data from both longterm trends and periods,codes and categorizes the longterm trends, and uses the category optimal value selection method to predict the music popular trends.Based on the TSMP algorithm, the Extend TSMP (ETSMP) algorithm is proposed, which is based on the subsequence pattern matching method and the additional processing of the newly released new album to achieve more accurate prediction.In the “2016 Chinese University Computer ContestBig Data Challenge Ali music popular trends prediction competition” hosted by Tsinghua University and Tianchi big data competition platform of Ali cloud, the participating team uses the proposed ETSMP algorithm to achieve good prediction for artist’s play times from september to october in 2016 and won the second place in this competition.
 

Key words: time series, music popular trends, category optimal value selection, subsequence pattern matching