Movie Box Office Prediction Based On The Stacking Method

With the continuous development and expansion of the film industry and the emergence of self-media,more and more researchers have begun to pay attention to how to use the data from the media to predict some social events.This article uses Douban data to predict movie box office as an example,making full use of movie information data and comment data on Douban platform to accurately predict the effect of movie box office,and has a good guiding role in the production of film producers and the marketing of movies.Based on the 150 films released throughout 2016 and January-November 2017,this article screens out 100 movies that have been released for a short period of two weeks as a data set for analysis.In view of the importance of predicting the factors that influence movie box office,and word of mouth has also become a factor that viewers value more and more,so this paper innovatively adds emotion feature factors and constructs movie reviews when selecting variables for movie box office prediction.Emotional tendency dictionary was used to extract positive emotion feature factor,negative emotion feature factor,and neutral emotion feature factor.The correlation coefficient between extracted factor and film box office reached 0.87,-0.89,and 0.78.Fully explained the importance of word-of-mouth predictions for movie box office,and also added three quantitative indicators for movie box office prediction.This article also improves the prediction model of movie box office.Due to the multiple types of data on the watercress web site,it is difficult to describe its complex relationship with a single model.Therefore,this paper has continuously adjusted and tried Stacking algorithm and based on multiple linear regression model and BP neural network model two base learners and SVM times.The learner constructs a 3-layer movie box office prediction model,and the calculated MAPE is 0.019,which is better than the similar model.Finally,the article also compares the prediction effect before and after the addition of emotional feature factors,and finds that adding the emotionalfeature factors of the review can significantly reduce the predicted MAPE,making the prediction model more accurate and more stable.