Tri-Data Approach to Movie Success: Historical Trends, Current Dynamics, and Sentiment Insights

Prediction of movie success has been a topic of interest for several years, and various methods have been proposed in research papers to achieve this. The main cause for this interest is the high risk involved in production, along with the high rewards that successful movies garner. The same holds true for various other serial media, including novels, comics, and advertisements. As the variables involved in making a media successful are numerous, a highly modular, chronologically comprehensive, and conducive architecture for inference generation has been proposed. The overall aim is to predict the success probability and simultaneously recognize the variable features that are subpar and those that are exemplary for various media forms and not just movies. The easy availability of quality and comprehensive data related to movies has led to the application of the architecture on movies in this paper for research. Through an extensive literature review, specific features were chosen that were observed to produce accurate predictions. The overall architecture model achieved an accuracy of 89.58%. Predictive analysis using historical data had the highest accuracy at 96.86%, followed by recent data at 91.03%, and current sentiment data at 87.64%. Quality targeted data might allow this architecture to be used for various serial media by utilizing different features and appropriate machine learning models for specific instances, made possible by the modularity of the architecture. The proposed architecture brings about novelty through its comprehensive nature, broader use cases, inference generation for better production of successive media and easy customizability.