AAI_2025_Capstone_Chronicles_Combined

Cinema Analytics and Prediction System

28

Figure 27: Genre classification result with multiple genres

The genre classification model development confirmed that data balancing and

architecture tuning are crucial for multi-label classification on imbalanced text data. While BERT

provides rich semantics, well-designed LSTMs can effectively model label co-occurrence. Future

work could add attention layers, fine-tune transformers on movie data, or explore generative

label prediction. The balanced LSTM offers a strong base for automated genre tagging and

related applications like cataloging and personalized browsing.

195

Internal