Emotion Trajectories in Cinematic Narratives: A Transformer-Based Analysis

This paper introduces a novel method for analyzing movie dialogues to predict characters’ primary emotions and the overall emotional tone of a film. Using natural language processing, we assess dialogues across emotional dimensions such as happiness, sadness, anger, fear, and more. Our approach involves collecting and preprocessing movie scripts, followed by building emotion classification models. By examining the context, sentiment, and linguistic features of dialogues, we predict character emotions and aggregate them to map the emotional trajectory of the movie. Diverging from traditional rating prediction models, our method aligns emotional experiences from movie scripts with reviewer feedback for improved accuracy. We also propose a sentimentbased movie recommender system that matches users’ emotions with film sentiments, offering a unique alternative to existing recommendation systems. Through extensive evaluation, our approach contributes to affective computing and provides deeper insights into the emotional structure of cinematic narratives.