The Prediction Of Music Popular Trend Based On Machine Learning

Relative policies were issued by the government to protect the music copyrights in 2015,and many opportunities were provided to the further development of music industry.It is very important for music studios and platforms to predict the trend of music in an accurate way and prejudge the music dark horses exactly.Machine learning technology has been adopted to solve the problems widely existing in data mining domains.Typical problems of music industry are studied in this presented thesis,and predicting models of popular music trends are constructed based on machine learning methods.The mainly work can be stated as follows:(1)Ranking lists are displayed on the music platform to reflect the popularity of music via the champion of monthly,weekly,daily.Two classic machine learning algorithms,artificial neural network(ANN)and support vector machine(SVM)are utilized to predict the play times of next month,next week and next day.ANN algorithm is utilized to do a large number of experiments focus on combined features,which are extracted by artificial rules,to find out the optimal feature selection scheme.SVM is utilized to do the same experiments based on the optimal characteristics;the results of MSE(mean square error)are 0.031,0.0033 and 0.0046.(2)Construct a combination model to predict the music popular trend,the model linearly combined ANN and SVM.The results of MSE are 0.027,0.0028 and 0.0040.Compared with the results of SVM,performance raised by 12.90%,15.15% and 12.90%.Experiments show that the single model have advantages,but a combination model can reduce the limitations of single model,avoid the risk of local optimal for ANN and fitting for SVM,and it can significantly improve the prediction ability.(3)The music popular trend prediction model based on machine learning only used data close to predicted target time range.This paper did experiments group by artists after clustering them.Experiments decomposed time sequence into time information granulation based on fuzzy set theory,constructed a “triangle” model,whose parameters including low,R,and up are predicted with SVM algorithm.And it can get accurate change range and change tendency in a short period of time.It plays an important guiding role for users which have original and using behavior and for the marketing activities of operators.The experimental results demonstrate that the predicting models on music development trend can provide much theory supporting and technology assistance in predicting the trends of pop music and prejudging the black horses of music artists,and can achieve outstanding commercial value.Compared with the traditional prediction methods,models based on machine learning not only have a perfect theory,but also have a strong robustness and generalization ability in practice,it can be directly transplanted into platform and be applied.