Enhancing Film Genre Classification Using FastText Embeddings, Bidirectional GRU (BiGRU), and Attention Mechanisms

This research aims to enhance the classification of film genres using advanced natural language processing techniques. By integrating FastText embeddings with Bidirectional Gated Recurrent Units (Bi-GRU) and attention mechanisms, the proposed model addresses the limitations of existing methods that struggle with capturing both local and global dependencies within textual data. The model's performance is evaluated on a dataset from IMDb, demonstrating its capability to predict film genres from textual descriptions accurately. Key contributions include the development of a robust model architecture that effectively handles out-of-vocabulary words and contextual nuances, implementing regularization techniques such as DropConnect to improve generalization, and using advanced embeddings to enhance semantic representation. The results indicate significant improvements in genre classification accuracy, particularly for frequent genres, showcasing the model's potential for practical applications in media content analysis. Future work will address data imbalance and explore more sophisticated architectures to enhance performanc.