AI-Assisted Automated Micro-film Script Generation and Analysis System

With the rise of short stories in the digital medium, there has been a growth of interest in developing intelligent, scalable systems that generate micro-film scripts. This paper introduces an Artificial Intelligence-assisted automated framework to generate micro-film scripts, and to analyze their narrative structure and content using a Fine-tuned GPT-2 Transformer Model. Movie scripts are pre-processed to normalize, tokenize by sub-words, and segment the text into fixed-length sequences to develop "screenplay" specific narrative patterns. The model is then further trained using the AdamW Optimiser with a learning rate of 5E-5, a batch size of 8 and training over 3 epochs; additionally, micro-film scripts will be generated using Nucleus Sampling (top-p = 0.9), and Temperature Scaling (temperature = 0.8). An extensive evaluation framework has been used to evaluate the outputs of the proposed framework via BLEU, ROUGE, Distinct-N, Semantic Similarity, Narrative Coherence, and Sentiment Consistency metrics. The experimental evaluations showed a proposed framework generated high semantic similarity (0.93), strong lexical diversity (Dist-2 = 0.83), and stable sentiment consistency (0.555) indicating a proposed framework generated fluent, creative scripts. However, global narrative coherence was evidenced to be only moderate (0.264); therefore, the proposed framework could provide a reproducible and scalable solution to allow for automated micro-film script creation and use in creative, educational and commercial industries.