Research On The Application Of Data Mining Technology In Music Trend Prediction

Accurate prediction of trends of music can enhance user experience,promote up the popularity of artists and coming artists,and increase profitability of a music platform.The prediction of such trends from historical data is of great significance for artists,music lovers and online music platforms.Data from online music platforms typically has high dimensionality,great complexity,and obvious characteristics of time series.Existing music trend models and traditional statistical models often struggle to effectively analyze the intricate relationship between artists,users,songs(downloads,playbacks,collections),and mine music data in depth.The shortcomings of these models motivates the research in this paper.In recent years,many forecasting problems have been studied from the perspective of regression forecasting and time series forecasting.Among them,regression forecasting model based on Random Forest and time series forecasting techniques have achieved promising results in many fields.To predict trends in music popularity,this paper constructs four models.The first is a regression prediction model based on the Random Forest algorithm,the second combination model of a Sudden Song Attenuation Model(SSAM)and the regression prediction model based on the Random Forest algorithm,the third a time series prediction model based on Long Short-Term Memory network and fourth an Auto-Regressive Integrated Moving Average time series model.The prediction effect of these models are then evaluated through experiments.The main work of this paper includes the following:(1)Research from the perspective of regression prediction,a model based on Random Forest algorithm is constructed to predict the trend in music popularity.Random forest techniques excel at handling the multi-dimensional characteristics of Internet music data and additionally it has fast training speed and strong anti-jamming ability.The process uses One-Hot encoding of user preferences,listening time habits and other multi-dimensional data,and then cluster several types of features after One-Hot encoding through K-Means.At the same time,aiming at the attenuation process of the special song which has a large number of playbacks recently,a Sudden Song Attenuation Model(SSAM)is proposed.The model,combining with the regression prediction model based on random forest algorithm and SSAM,makes a great improvement compared with original forecasting model.(2)Research from the perspective of time series prediction,a music trend prediction model based on Long Short-Term Memory(LSTM)network time series is constructed and evaluated.In view of the obvious time series characteristics of online music data,the LSTM network is suitable for time series modelling and prediction and it has the advantage of remembering long-term information.It is specifically designed to solve the problem of long-term memory.The prediction ability of the model is also tested through experiments.In addition,Constructing an Auto-Regressive Integrated Moving Average time series prediction model of music trends.The Auto-Regressive Integrated Moving Average model(ARIMA)helps reveal insights in most time series data.Different from LSTM,the differential ARIMA is suitable for short-term prediction research.The model has the advantages of simplicity,only requiring endogenous variables,wide application range and small prediction error.The prediction effect of the model is also tested by experiments,and the experimental results are compared with the LSTM time series prediction model.The core evaluation index F has been defined.The four popular trend prediction models in this paper are tested by using the Internet music data of the Ali.The evaluation index F values of the four models are compared vertically to analyze the quality of the prediction.The experimental results show that the music trend prediction model based on Random Forest algorithm has a higher F value and better prediction performance than the two time series models.Combining the sudden songs attenuation model in the viewing effects of music with the random forest increases the F value and the overall predictive ability.The experimental results demonstrate the paper is effective in predicting the trends of music and achieves the goal of prediction.