Deep Learning over 20 Years of Cinema Ticket Sales

Abstract

For many years, movie theaters across Norway have been manually setting up its weekly movie schedule according to the prior knowledge of the operators. This is a slow and laborious method that in this day and age is outdated and should be automated.

This thesis explores how this process can be automated using deep learning. By exploring 20 years of sales data we gain insight into what can cause screenings to have a high or low coverage. We use this insight to assist with feature selection for the neural network, which in this thesis will be the TabularModel from fastai. With this model we are able to predict the coverage of a future screening.

Our objective is to use these predictions to create a suggested schedule that will optimize ticket sales for movie theaters.