Using sentiment and social network analyses to predict opening-movie box-office success

Abstract

In this thesis, we explore notions of collective intelligence in the form of web metrics, social network analysis and sentiment analysis to predict the box-office income of movies. Successful prediction techniques would be advantageous for those in the movie industry to gauge their likely return and adjust pre- and post-release marketing efforts. Additionally, the approaches in this thesis may also be applied to other markets for prediction as well. We explore several modeling approaches to predict performance on the Hollywood Stock Exchange (HSX) prediction market as well as overall gross income. Some models use only a single movie's data to predict its future success, while other models build from the data of all the movies together. The most successful model presented in this thesis improves on HSX and provides high correlations/low predictive error on both HSX delist prices as well as the final gross income of the movies. We also provide insights for future work to build on this thesis to potentially uncover movies that perform exceptionally poorly or exceptionally well.