AAI_2025_Capstone_Chronicles_Combined

Cinema Analytics and Prediction System

13

movie revenue using a broad set of features including cast, release date, and text-based

overviews.

Model Architecture, Training, and Optimization

This section describes the architectural design, training and optimization procedures for

all components of the project. While each task addresses a different machine learning

objective, NLP based models share many foundational components, including textual feature

engineering, embedding strategies, and hybrid model design, though they address different

machine learning tasks: similarity ranking and multi-label classification. Before model training, a

unified preprocessing pipeline (Figure 8) was applied across all NLP-based tasks:

Figure 8: NLP pre-processing pipeline

Movie Recommendation Models

The recommendation system was designed as a content-based filtering engine using

three different modeling approaches. Each model used cosine similarity to compare vector

representations of movies and rank similar titles.

180

Internal