Box Office Forecasting Base On Weibo Data

From a global perspective, the movie industry is the fastest-growing industries. Thousands of picture movie is released every year, the CAGR of movie market is growing year by year. However, people's preferences is very difficult to predict, it is a very high risk to meeting the self-financing balance after the movie was released. So the film industry has a great concerning to forecasting movie box office.After “audience research” and abandoning the method depending on the meta data to resolve the problems of model improvements and impact factor, The research has fully used the user-generated content to establish prediction models, including the blogger model, Twitter model, Google model and Wikipedia model. By definition, forecasting models are more valuable if he forecasts are made as early as possible. In other words, a forecast that is extraordinarily accurate is not useful if t is made only a few days before the closing. However, there is an information discrepancy between the periods prior to release and after release. When a motion picture has not yet been released, only a few pieces of information in weibo. Most of data were released in the first week or day after the motion picture released.so we can't get a good predicting accuracy.In order to eliminate the influence of predicting result for weibo data's time-lag, this paper proposes a new forecasting model based on weibo data to predict the box office, we put forward the whole forecasting process is divided into consecutive three models, which use predictive algorithm. that is(1) prior to,(2) a week after, and(3) two weeks after release after and to predict the box office after the movie was released in one week, the cumulative box office of two weeks after the film released, the cumulative box office of three weeks after the film released. The model variables are divided into two categories, one is the actual box office data, and second is microblogging data. When we predict the box office of two or three weeks the film released, we have get the first and second week of real box office, it can also be added to the box office prediction model. The Weibo data are divided into four categories, mention, emotion mention, positive emotions, negative emotions and derived 12 variables. Three prediction models are optimized by the genetic algorithm support vector regression. We get the optimal parameters of each model by simulation experiment and we prove the scientific and reasonable of the predicting model by measuring the MAPE of model, more later the predicting time,more higher the accurate rate.