Prediction Research On Movie Box-office Based On Stacking Ensemble Learning

In recent years,with the rise of consumption level and the improvement of film technology,the film industry has developed rapidly,and film consumption has become a new fashion.At the same time,with the rise of social networks,when viewers choose movies,they are no longer solely based on their preferences,and will be influenced by movie reviews to a greater extent.To explore the influence of movie reviews on the film box-office,to deal with the complex and difficult quantification of the influencing factors and the low classification accuracy of the existing prediction models,the paper improves the traditional box-office prediction model,introduces the emotional characteristics of film reviews,and puts forward the box-office prediction model based on Stacking integrated learning,and then improve the accuracy of box-office prediction.The paper combs the domestic and foreign box-office research literature,studies the influence of online reviews on box-office and the improvement of film prediction model.The paper deeply excavates the information of film reviews and conducts sentiment analysis to explore the influence of emotional characteristics of film reviews on film box-office.And through constructing the integration model of film Stacking algorithm to improve the existing model.The paper mainly carries on the following three parts of work.(1)Constructing the sentiment analysis model of film review text.The paper builds Bi LSTM sentiment analysis model for labeled film review texts,compares and evaluates it with LSTM and GRU models,the result shows that the Bi LSTM model has a better training effect and calculates the sentiment score of the required film review text.(2)Screening influencing factors to build box-office model.The paper selects the boxoffice data of the top 100 films and the relevant influencing factors from 2017 to 2019,cleans and quantifies influencing factors,analyzes the influence of each factor on the box-office,and selects the main influencing factors through the importance feature graph of XGBoost algorithm,finds that the box-office of the first week and the emotional score of film reviews have a great impact on the box-office.By using six kinds of algorithm of machine learning to build box-office forecasting model,through the grid search and cross validation parameters optimization to further improve the prediction accuracy of the model.(3)Constructing the Stacking integrated learning film box-office prediction fusion model.The paper evaluates the model performance according to the root mean square error comparison,uses the Stacking integrated learning method to fuse models with better fit,compares with the emotional features before and after the introduction and fusion of different based learning device model prediction effect,and compared with XGBoost model,the fusion model has better prediction effect.