To what extent data mining of social media can be applied to anticipate cinematic incomes?

Abstract

For the last fifteen years, Data Mining has taken a fast-growing importance within thegeneral process of decision-making in the Business world. Extracting useful information from large datasets in order to convert them in a powerful tool of decision/prediction is obviously used for many different applications. This phenomenon mustbe related with the amazing growth of available data (particularly through the Internet and theNew Social Media).Most of the companies, skeptical at the beginning, are now getting the fabulous scope ofthis wide resource. However, Data Mining must be used with precautions and can’t replace anyhuman judgment. This last point is crucial to understand the position that the cinematicindustry decided to adopt in front of such a tool. Indeed, cinematic industry copes with apermanent dilemma which is inherent to its own definition: can we consider the filmmaking/selling process as an industrial activity or only as an artistic process. Even if the commercial part of a motion picture (production, international sales, distribution) is easilycomparable to any other commercial process, cinematic actors (from the film makers to thedistributor) are reluctant to assimilate their own activity to any other kind of commercial serviceaiming at the consumers’ society. “You’ll never sell a movie as you could sell any tooth brush!”This position is justified by the cultural value of any artistic outcome which is hardly definableand predictable. This distrust and skepticism in front of any statistical or data mining approachis also understandable through the constitution of the industry itself. At each level of the film making/selling process, several persons are in charge of the artistic and commercial evaluation of the film. Accepting the fact that computer sciences are more able than any person to predictthe success of a film would signify the uselessness of these current key players.In addition, most of the recent studies (focused on the prediction of cinematic resultsthrough data mining techniques) coped with this famous and inherent problem to forecast accurately the results of a cultural product which partially differs from any other kind of“needed consumption” (foods, cars, etc). The accuracy of these studies was most of the time questionable. First, we will try to demonstrate that Data Mining, particularly at the age of social media,can find a very useful application within this industry. Then we will analyze and develop severalmodels to predict the potential revenues/success of a movie on the market (mid-term prospect). Eventually, we will conclude discussing and underlining briefly the various limits andimprovements peculiar to the cultural product.The first problem we faced was related to the diversity of consumption modes due tocultural divergences. The way of “consuming” a film is totally different between India and theUS for instance. Because of these divergences, we took the decision to focus our study only onthe French market where we were able to collect easily information and data. However, thisstudy will be also the opportunity to make a comparison between the various cultural andgeographical ways to “consume” a movie.