Applying Machine Learning to Predict Film Daily Audience Data: System and Dataset

Audience data is highly correlated with the levels of film viewership, as they can impact on people’s perceptions of films, resulting in whether or not they would watch a film or even recommend to others. Hence, if audience data can be properly analyzed, they can provide important clue to help predicting the trend of daily film statistics, such as box office, attendance rate, etc., which would further help cinemas to make wise marketing decisions. Motivated by this, we propose a novel audience data prediction system based on the recent advance of deep learning. Our approach begins with applying Fourier Transform-based algorithm to encode multi-channel time-series audience data into a set of feature maps. Then, these feature maps are fed to Generative Adversarial Networks (GANs) to predict and generate future audience data. To evaluate the proposed approach, we collected a dataset consisting of 200 films across three years (2017, 2018 and 2019), where 15 different daily attributes of 30 days are provided for each film. To help potential research of other researchers, we made it available online. The experiment results illustrated the superior performance of our algorithm in comparison to the baseline.