A Machine Learning Model for Predicting a Movie Sequel's Revenue Based on the Sentiment Analysis of Consumers' Reviews.

The relationship between the performance of movie sequels, the performance of the corresponding original movies and the users’ review sentiments is actively studied in the scientific community. However, the precise constitution of this relationship remains unclear due to the complex multidimensional nature of the problem. In particular, the precise correspondence between the users’ review sentiments and the topic structure of the reviews (that represents the aspects of the movie that impacted the sentiment the most) is yet to be fully understood. In this study, a machine learning topic modeling algorithm (Latent Dirichlet Analysis, LDA) is performed on the three movies from the Jurassic World franchise. The analysis is performed on a dataset of reviews gathered from the IMDB website. The reviews are separated into six datasets – a positive and a negative subset for each of the three movies. The outputs of the topic modeling are represented as word clouds of the most salient terms. The subsequent analysis of the word clouds demonstrates the heterogeneity of the topics within reviews and the nature of the ambiguity that often complicates the vocabulary-based sentiment analysis. Based on the results of the topic modeling, using comparative methods we determine the possible reasons behind the significant decline of the box office performance for “Jurassic World: Dominion” and the franchise in general. Our result illustrated that successful sequel would have to be consistent with other movies of the franchise and to have enough originality at the same time to receive positive feedback. Future works includes developing an approach that can leverage the heterogeneity of the LDA-produced topic representations, applying roBERTa model to handle sentimental analysis, and predicting movie sequel’s revenue based on machine learning models.