Application And Research Of Box-office Revenue Prediction Of Movie Based On Functional Network

Film as a booming emerging industry, more and more individuals or groups show great interest in movies investment. At present, in the film markets, most of profitable movies were produced by private enterprises. Obviously, profit maximization is the ultimate goal for all investors, especially these private enterprises. However, there are many uncertain factors affect the film market, and no so-called ‘golden rule’ suitable for the whole film industry. How to improve the rate of return is an urgent task lies before all film practitioners.In the pre-production staging, the prediction of box-office revenue is an indispensable part, which can help investors ensure return on investment(ROI) and manage risks. Based on the accurate prediction, the film investors and issuers can utilize the cost reasonably during film production and adjust the marketing strategy appropriately in the period of releasing.As a new subject was derived from neural network, functional network(FN) rising in recent years and widely used in various fields of social life. Currently, whether at home or aboard, there are few researches on the prediction of film box-office revenue using functional network. In light of this situation, we attempt to apply function network to predict the movie box-office. Our analytic works in concrete are mainly as follows:(1) In this paper, firstly, according to the present situation of Chinese movie market combined with previous studies on the prediction of box-office, we selected some impact factors which have effects to the box-office revenue. Secondly, initial data associated with these impact factors was extracted from the special websites by Jsoup technology. Finally, we obtained the effective training samples based on the cleaning and integrating of initial data.(2) Sensitivity analysis was carried out on the above impact factors, and then ranked parameter sensitivity using Morris method. The last input variables of model determined by their effects on the model outputs.(3) Established the FN-based prediction model of film box-office revenue. First of all, in order to make the training set more accurate and effective, clustering analysis was conducted on the training set. Then, the topological structures and primary functions of model were determined, and the parameters of the model were obtained by Gaussian elimination method. At last, built prediction model and was further validated by the independent test set.(4) The simulated tests manifested that the FN-based method is significantly superior to the well-known BP neural network method, which has a smaller average relative error and more rational root-mean-square error(RMSE). Moreover, compared to the BP neural network models, our model designed with a more optimized network structure while less computing time.In conclusion, this topic researches would provide scientific reasonable decision supports for the film investors and issuers. Last but not least, our studies have reference value and guidance significance to the popularization and application of functional network in the engineering practice.