The Research Of Movie Box Office Prediction Based On Social Media Data

With the rapid development of film industry,film production is increasing gradually.Film is a high-risk industry.In addition to being influenced by such factors as the movie plot,the quality of the filming,the release time and the amount of filming,the film box office is also related to advance publicity,word-of-mouth,movie stars,director and other factors.We can adjust the early capital investment,such as the advance publicity of the movie,and reduce investment risk through the movie box office forecast.Social networking platforms,especially Sina Weibo,as a nascent social media,have a large number of active users,including many movie actors and directors.It is of great commercial value to make the box office forecast by mining the data related to the theme of the movie in the social network.In the current research on box office prediction,it is mainly through the analysis of the basic properties of the film to predict,ignoring the importance of actors in film box office prediction.Based on the above problems,we crawled the tweet and popular comments of movie actors and actresses on Sina Weibo platform,and used the basic information of movies in Douban movie as the forecasting data.Then we extracted the features of purchase intention,popularity of actor,emotional tendency and topic distribution based on popularity,and predicted the box office using a variety of regression algorithms.This article focuses on two methods of feature extraction:the first one is the extraction of emotion features based on CHI-PCA.The main method of emotion classification in box office prediction is based on the classification method of sentiment lexicon.The classification result of this method depends on the quality of sentiment lexicon.Therefore,we used the method based on the statistics of sentiment classification.And merged it with the dimensionality reduction of PCA to preserve the integrity of features.The method does not produce redundant features that affect the classification effect at the same time.The second one is the LDA topic distribution features based on popularity.This feature puts forward the calculation method of tweet popularity and word popularity.Based on the traditional LDA model,tweet popularity is merged to obtain the topic distribution based on popularity.The experimental results show that the two characteristics can effectively improve the box office forecasting results.