Advancing predictive content analysis: a natural language processing and machine learning approach to television script data

Abstract This study introduces a predictive framework for estimating television episode viewership using machine learning and natural language processing applied to over 25,000 TV scripts. By analyzing linguistic and emotional features embedded in dialogue, the research identifies content patterns linked to audience viewership. Multiple regression models, including OLS, Lasso, Ridge, Elastic Net, Gradient Boosting, and XGBoost, are trained to forecast next-episode viewership, explaining up to 50% of variance at the genre level and 41% at the series level. These findings suggest that early-stage script analysis can offer actionable insights for media development and marketing teams. Rather than viewing scripts solely as creative artifacts, this research highlights their potential as data assets for content strategy, allowing for more informed decisions in greenlighting, promotion, and brand alignment.