PhD Dissertation Defense at the College of Information Technology, University of Babylon on Automated Movie Genre Classification and Age Suitability Using Artificial Intelligence Techniques

Duhaa Fadill Abbas
As part of its ongoing scientific efforts to support advanced research and keep pace with developments in artificial intelligence and data analytics, the College of Information Technology at the University of Babylon witnessed the defense of a PhD dissertation in the Software Department by Mr. Yaseen Khudair Abbas, entitled “A Proposed Automated Approach for Movie Genre Classification and Age Suitability Based on Film Scripts.” The dissertation was supervised by Professor Dr. Ahmed Habib Al-Azzawi, and attended by a number of faculty members, researchers, and postgraduate students.
The dissertation addressed a significant topic in multimedia content personalization, as movie genre classification and age suitability assessment represent fundamental components in organizing digital media content, particularly in light of the rapid expansion of online film libraries. It also highlighted the importance of widely adopted rating systems, such as the Motion Picture Association of America (MPAA), in guiding audiences—especially parents—in selecting appropriate content and filtering out unsuitable material.
The research aimed to develop a unified model capable of automatically predicting both movie genres and age ratings based on full-length film scripts, thereby overcoming the limitations of previous studies that relied on summaries or partial data. It also addressed challenges associated with traditional models, including restricted input lengths and limited contextual understanding of long textual narratives.
The proposed methodology was structured into two main stages. The first stage focused on multi-label genre classification using a multimodal approach that integrates textual and visual information. This was achieved by employing advanced models such as RoBERTa-large with a segmentation strategy to process long texts effectively, combined with visual features extracted from movie posters using EfficientNet. To enhance performance in the presence of class imbalance and overlapping genres, an asymmetric loss function was utilized during training.
The second stage involved predicting MPAA ratings using textual data alongside sentiment-based features extracted from film scripts. The results demonstrated that the proposed model outperformed several baseline deep learning approaches, including LSTM and Bi-LSTM models, achieving strong performance metrics in genre classification as well as high accuracy in age rating prediction.
This dissertation contributes to the development of more accurate content recommendation systems and parental guidance tools by enhancing automated understanding of film narratives and their suitability for different age groups. Moreover, it provides a scalable and reproducible framework for future research in multimodal machine learning, demonstrating how long-form textual narratives and embedded emotional features can be effectively leveraged in classification tasks across various domains.