Research for Music Trends Prediction Based on LSTM-RPA

doi:10.3778/j.issn.1002-8331.2105-0464

Abstract

Abstract: The big data about music history contains information about time and users’ behavior. By analyzing the behavior data of artists and listeners, researchers can predict the trend of music play-count, and then predict the trend of popular songs accurately. The traditional trend prediction models can predict the short trend better than the long trend. However, the traditional trend prediction model is difficult to obtain good results because of the serious attenuation of historical information in the long trend prediction. This paper proposes an improved LSTM rolling prediction algorithm（LSTM-RPA） for the loss of historical information in the long trend prediction. RPA combines LSTM model historical input with current prediction results as model input for next time prediction, so that the historical information can flow along the prediction trend. The experiment uses the competition data set of “2016 China Collegiate Computing Contest—Big Data Challenge：Popular Music Prediction”. And the experimental evaluation is F value provided by the competition organizer. The evaluation results show that the LSTM-RPA model has increased F score by 13.03%, 16.74%, 11.91% and 18.52%, has decreased the mean error by 39.02%, 48.55%, 36.02% and 52.88% under the same conditions to predict the music play-count by artists in the next 30 days, compared with LSTM, BiLSTM, GRU and RNN. And proposed method outperforms traditional sequence models, ARIMA, F score is increased by 10.67% and mean error is decreased by 32.64%.

Key words: long trend prediction, long short-term memory（LSTM）, music popularity prediction, time series

摘要： 大数据背景下的音乐历史数据蕴含丰富的时间信息和用户行为信息，通过分析音乐艺人和听众行为数据，可以较为精准地预测音乐播放量走势，进而预测音乐流行趋势。传统的时间序列预测模型可以准确预测短趋势，但在长趋势预测中受历史信息衰减的影响，难以获得较好的效果。针对LSTM在音乐长趋势预测中历史信息衰减的问题，提出改进的LSTM滚动预测模型，该模型在预测阶段将前一次输入与当前预测结果相结合，使得历史信息可以沿预测趋势方向流动，从而缓解模型在长趋势预测中的历史信息衰减。实验采用“2016中国高校计算机大赛——大数据挑战赛：阿里音乐流行趋势预测”的比赛数据集，并使用比赛主办方提供的F值进行评估。实验结果显示：在相同条件下预测艺人未来30天的每日音乐播放量，最优LSTM滚动预测模型与LSTM、BiLSTM、GRU、RNN相比F值提高13.03%、16.74%、11.91%、18.52%，平均误差减少39.02%、48.55%、36.02%、52.88%；与传统的时间序列预测模型差分整合滑动平均自回归模型相比F值提高10.67%，平均误差降低32.64%。

关键词: 长趋势预测, 长短期记忆网络（LSTM）, 音乐流行趋势预测, 时间序列

LI Kun, LI Meng, LI Yanling, LIN Min. Research for Music Trends Prediction Based on LSTM-RPA[J]. Computer Engineering and Applications, 2022, 58(24): 134-142.

李堃, 李猛, 李艳玲, 林民. 基于LSTM-RPA音乐流行趋势预测研究[J]. 计算机工程与应用, 2022, 58(24): 134-142.

References

[1] CHENG Z，SHEN J，ZHU L，et al.Exploiting music play sequence for music recommendation[C]//International Joint Conference on Artificial Intelligence，2017：3654-3660.
[2] LEE J，LEE J S.Predicting music popularity patterns based on musical complexity and early stage popularity[C]//Proceedings of the Third Edition Workshop on Speech，Language & Audio in Multimedia，2015：3-6.
[3] REN J，KAUFFMAN R J.Understanding music track popularity in a social network[C]//Proceedings of the 25th European Conference on Information Systems ECIS，Guimar?es，Portugal，June 5，2017：374-388.
[4] VALIPOUR M，BANIHABIB M E，BEHBAHANI S M.Parameters estimate of autoregressive moving average and autoregressive integrated moving average models and compare their ability for inflow forecasting[J].Journal of Mathematics and Statistics，2012，8（3）.
[5] LEE Y S，TONG L I.Forecasting time series using a methodology based on autoregressive integrated moving average and genetic programming[J].Knowledge-Based Systems，2010，24（1）.
[6] SALGANIK M J，DODDS P S，WATTS D J.Experimental study of inequality and unpredictability in an artificial cultural market[J].Science，2006，311（5762）：854-856.
[7] 吕倩倩.基于机器学习的音乐流行趋势预测[D].兰州：兰州大学，2017.
LYU Q Q.The prediction of music popularity trend based on machine learning[D].Lanzhou：Lanzhou University，2017.
[8] 颜家康.BP神经网络算法在音乐流行趋势预测中的应用研究[D].成都：西南交通大学，2017.
YAN J K.Research on the application of BP neural network algorithm in the prediction of music popular trend[D].Chengdu：Southwest Jiaotong University，2017.
[9] 郁伟生，邓伟，张瑶，等.基于时间序列的音乐流行趋势预测研究[J].计算机工程与科学，2018，40（9）：1703-1709.
YU W S，DENG W，ZHANG Y，et al.Music popular trends prediction based on time series[J].Computer Engineering and Science，2018，40（9）：1703-1709.
[10] HOCHREITER S，SCHMIDHUBER J.Long short-term memory[J].Neural Computation，1997，9（8）：1735-1780.
[11] YU H，LI Y，ZHANG S，et al.Popularity prediction for artists based on user songs dataset[C]//Proceedings of the 2019 5th International Conference on Computing and Artificial Intelligence，2019：17-24.
[12] 王振业，叶成绪，王文韬，等.基于LSTM-Att方法的音乐流行趋势预测[J].计算机技术与发展，2020，30（9）：188-193.
WANG Z Y，YE C X，WANG W T，et al.Music trend forecast based on LSTM-Att method[J].Computer Technology and Development，2020，30（9）：188-193.
[13] LAUREN S，HARLILI S D.Stock trend prediction using simple moving average supported by news classification[C]//2014 International Conference of Advanced Informatics：Concept，Theory and Application，2014：135-139.
[14] 徐先峰，刘阿慧，陈雨露，等.基于气象因素充分挖掘的BiLSTM光伏发电短期功率预测[J].计算机系统应用，2020，29（7）：205-211.
XU X F，LIU A H，CHEN Y L，et al.BiLSTM short-term forecasting method for photovoltaic power generation based on full exploiting meteorological factors[J].Computer System Application，2020，29（7）：205-211.
[15] 张金磊，罗玉玲，付强.基于门控循环单元神经网络的金融时间序列预测[J].广西师范大学学报（自然科学版），2019，37（2）：82-89.
ZHANG J L，LUO Y L，FU Q.Predicting financial time series based on gated recurrent unit neural network[J].Journal of Guangxi Normal University（Natural Science Edition），2019，37（2）：82-89.