AAI_2025_Capstone_Chronicles_Combined

Cinema Analytics and Prediction System

15

Figure 9: Text and structured features

Figure 10: LSTM architecture

These learned embeddings captured both narrative and numerical dimensions of

similarity, enabling more refined recommendations (e.g., thematically similar and highly rated

movies). Cosine similarity was applied to the penultimate-layer outputs to generate ranked

suggestions.

Genre Classification Models

The genre classification system framed the problem as a multi-label classification task,

where a movie could be associated with multiple genres simultaneously. The system reused

much of the recommendation model’s infrastructure while adapting outputs and l oss functions

accordingly.

182

Internal