Narrative-Based Script Quality Prediction Using Transformers

Bys: D Banu Kranthi and Matsa Vasavi

Traditional approaches often rely on subjective judgment and historical experience, which may not accurately estimate audience reception. To address this limitation, the proposed system combines Natural Language Processing (NLP) and structured metadata analysis to evaluate a drama before release. DistilBERT is used to extract meaningful contextual information from drama summaries and screenplay descriptions, while metadata features such as genre, number of episodes, number of seasons, cast tier, production tier, adaptation status, lead actor popularity, and screenwriter performance are incorporated to enhance prediction accuracy. Feature engineering techniques and class-weighted learning are applied to improve model performance and handle class imbalance. The extracted textual and metadata features are fused and processed through a neural network classifier to categorize dramas into Hit, Average, or Flop classes. The system provides an early assessment of a drama's potential success, assisting producers, writers, and production teams in making informed pre-production decisions.

• Format: Paperback
• Language: English
• ISBN: 6630557662
• ISBN13: 9786630557664
• Release Date: October 2026
• Publisher: LAP Lambert Academic Publishing
• Length: 52 Pages
• Weight: 0.19 lbs.
• Dimensions: 9.0" x 0.1" x 6.0"