Temporal and Lexicon-Aware VADER-based Sentiment Modeling for Box Office Revenue Prediction

There are several factors that affect the box office performance of a movie. The success of a movie may depend on the star cast, acting, direction, promotion, and marketing. This study presents a method based on an enhanced VADER (Valence Aware Dictionary and sEntiment Reasoner) model to predict the box office performance by analysing the sentiment of movie related news headlines. VADER can effectively analyse the sentiment of a general text, but it cannot capture domain-specific sentiments. It also fails to capture the temporal trends in the data. To overcome these limitations, we introduce three major enhancements: (1) a temporal sentiment profiling technique that can capture the changes in sentiments over time (2) a weighted sentiment based on closeness to the movie's release (3) adding relevant words to VADER's lexicon that appear in movie headlines. These values are then used to train various Machine Learning models to predict the box office performance. A comparison of VADER-only and enhanced VADER-based approach of around 5000 headlines and box office revenues shows that the enhanced model performs better than the baseline VADER-only approach. Among all Machine Learning models, RF regression achieved the best R2 score of .77.