Using reviews on the web to predict box office success with machine learning methods

Abstract

Movie production is a serious undertaking that requires a significant amount of investment. The success of a movie, i.e., positive return on investment, can only be realized with a good performance at the box-office well after the movie’s release at theaters. Ability to forecast box-office revenues, which is the amount of money a particular film generates, before a movie’s release can decrease the financial risk of film producers. Movie distributors or movie theater organizers can make better decisions about the time and space they assign for a specific movie if they have accurate predictions. However, the box-office success of a movie relies on many factors, some of which are highly subjective, making accurate predictions a challenging task. Before movies are released in theaters, their trailers, or previews, are made public by movie producers, for which the online social platforms such as YouTube are increas ingly utilized as the distribution medium. Such platforms also provide the opportunity for the users to react to and comment on them. In this thesis, we attempt at using user feedback on movie trailers on YouTube as additional features for box-office success prediction of movies with machine learning. Our results indicate that people’s reactions to movie trailers provides a set of helpful features in making more accurate predictions on movie box-office success.