Mining Emotion In Box Office Prediction Based On Social Network

In recent years,film art has been made accessible to everyone.The film market is full of fierce competition every year.Some films have enough superior conditions for talent showing itself while others still get poor box office results even though they hired high-traffic stars.It has to be said that the overall box office market is not satisfactory during the last two years.Contradictions between film producers and film review media are placed in front of the public.It seems that some films fail to achieve expected grades.The direct reason is the review from audience.Some movie scoring websites,film media and audiences themselves quickly occupy an important position in the public opinion.This paper focus on quantifying the impact of film reviews on the film market and how to mining valuable information in order to predict the box office.On the one hand,this paper introduces multi relational topic model to excavates concerns among movie.On the other hand,it trains a LSTM model to uncover the emotional feature within short film reviews.Finally,it proposed a muti-satge box office prediction model with the conclusion that a non-linear SVR model fit first week well and the Lasso model is suitable for next weeks.Sepecifically,the work of this paper mainly include the following aspects.(1)Extract opinion topics which is composed of specific words in reviews.After filtering nonthematic reviews,the remaining topics shows the focus of audience in respective movie.A multirelational topic model MRTM has been proposed in this paper.The experiments shows that MRTM effectively improves the quality of topics in short movie reviews.(2)The traditional analysis of factors in box office prediction includes word-of-mouth and actors or direrctors.This paper incorporates emotional characteristics to traditional features by LSTM-based sentiment model.It enriches the external features of the prediction model.By analyzing the changes of audience’s emotional,it provides strong support for the accurate box office prediction.(3)Based on the analysis of the film life cycle,a multi-stage box office prediction model has been designed in this paper.It provides a method to measure the dynamic impact of the moviemakers.Expriemnts show that it is more reasonable using different prediction models at different stages during movie onscreen.Specifically,the non-linear SVR model fits the first week well and the Lasso model is used to predict box office in the following weeks.