Model-based Predictive Control for Continuous Success Planning in Movie Production

The movie industry across the globe is a multi-billion-dollar business for stakeholders. Perhaps the single most financially dominating recipe among all flavors of media entertainments available to the audience. A plethora of carefully coordinated onerous efforts encompassing acting, direction, scriptwriting, casting, editing, and production goes into making and releasing a movie. Financiers and producers reel through several financial losses if the movie does not perform well at the box office. Given the complicated composition of time, money, imagination, creativity, and risk, it becomes imperative to have some measurable, quantifiable, and controllable parameters to predict movie performance on box-office during various stages of making through leveraging insights from plot-summaries, trailers, posters, teasers, social-buzz, cast/crew-selection etc. — Can specialized Artificial Intelligence techniques help from historical data across these modalities to best predict the success of a movie and be the foundation for creating a sustainable architecture for Model-based predictive control? This paper proposes a Model-based predictive control to create a closed-loop feedback system that will enable superior production planning. Model-based predictive control has seen success in the Industrial automation domain and has been historically used for establishing feedback control loops for industrial automation processes. The paper also presents experimental results on the foundation needed for a model-based predictive control: A hybrid, yet comprehensive RNN/ LSTM/CNN-based neural network architecture that can predict the success of the content much earlier to optimize production cost. It leads to a useful foundation that enables content creators to manage especially the pre-production part of the journey much better by making the right decisions that enhance the probability of success.

Published

2021-11

Content type

Original Research

Keywords

CNN, LSTM, RNN, Movies, Posters, Videos, Genres, Explainability, SHAP, Grad-CAM, Multimodal architecture, Model-based predictive control

DOI

10.5594/M001949