Research On Combination Forecast Model Of Variety Show On Demand

The emergence of Internet and electronic devices has made people's entertainment lifestyles rich and diverse.In recent years,in addition to movies and TV dramas,many excellent variety shows have appeared in the public's view,and the variety show industry has also developed rapidly.The mode of overseas variety shows has inspired the production of domestic variety shows.Not only did well-produced TV variety shows appear in China,but also a large number of online variety shows have emerged after 2014.There are more and more viewers watching variety shows on mobile devices.The amount of on-demand is becoming more and more important in measuring the success and popularity of a variety show.Therefore,the prediction of the amount of variety shows on-demand can not only reflect the viewers' preference,but also reflect the commercial value of variety shows,which can provide the basis and reference for the copyright purchase,advertising pricing and investment decisions of advertisers.This paper selects 293 variety shows in Tencent video which first broadcast time is from2017 to 2020 and constructs the influencing factors index system and prediction models of their on-demand.First of all,by reading and sorting out the literature,19 factors from the perspectives of related factors of the program and network data are selected to construct the index system of the factors influencing the on-demand volume of variety show.Secondly,this paper processes the collected data as filling missing data with values,converting different variables into dummy variable and normalizing them,and makes a simple descriptive statistics to help us to understand the characteristics of sample data,which prepares for empirical analysis.Then eight main influencing factors are filtered out by stepwise weighted regression,and a multiple stepwise weighted regression model is constructed to predict the demand volume of 10 variety shows.At the same time,the random forest algorithm is used to train the sample data and calculate the importance score of each variable.The top six variables are selected to build a model to predict the on-demand volume of variety shows.Finally,three kinds of weight coefficients are used to construct combined models to predict the on-demand volume of variety shows.They are the MAPE weight coefficient,the least square method weight coefficient and their combination weight coefficient.The accuracy of the prediction results of each model is evaluated by the MAPE.The empirical analysis results show that the accuracy of the prediction results of each model from high to low is the combination model of combination weight coefficient,the combination model of MAPE weight coefficient,the combination model of least square method weight coefficient,the random forest model and the multiple stepwise weighted regression model.Among them,the combination model using combination weight coefficient has the highest prediction accuracy,and the MAPE is only 0.0858.The multiple stepwise weighted regression model has the largest error that is 0.2442 of the prediction results,which is 184.62% higher than the combined model.The MAPE of random forest model is 0.0967.It can be seen that the prediction effect of the random forest model is better than the multiple stepwise weighted regression model in the single models.And the prediction effect of the combination models is better than the single models.