Document Type : Research Paper

Authors

• Zohreh Jafarbeglou
• MohammadAli AfsharKazemi ¹
• soheila jokar ²

¹ Department of Industrial Management, Central Tehran Branch, Islamic Azad University, Tehran, Iran

² Department of Mathematics, Islamic Azad University, Qeshm Branch, Qeshm, Iran

Abstract

Abstract
This research designed and implemented an intelligent multilingual BERT-based model for detecting and analyzing emotional and behavioral patterns in Persian and multilingual screenplays. The study employed an experimental method, utilizing a dataset comprising 35,800 text samples from 1,700 Persian and English screenplays, with targeted sampling for each emotional class. Linguistics and psychology experts annotated the data into eight emotional classes based on Plutchik's (1980) model. The process involved data collection from cinematic archives, preprocessing (including cleaning, normalization, and tokenization), dataset preparation, and supervised learning with fine-tuning of the model. The findings demonstrated the model's high capability in accurately identifying characters' emotions and behaviors, achieving an overall accuracy of over 98.22% and an average F1-score of approximately 98%. The model provides a practical tool for improving emotional coherence and reducing rewriting costs in screenwriting. While challenges remain in distinguishing semantically overlapping emotions, this study offers a solid foundation for future advancements in intelligent narrative analysis.

Keywords

• Keywords: Emotional analysis
• Intelligent assistant
• Multilingual BERT
• Persian screenplays
• Character behavior

Main Subjects

• Data, information and knowledge management in the field of smart business