AAI_2025_Capstone_Chronicles_Combined

Cinema Analytics and Prediction System

10

Amazon Prime Video often implement hybrid recommendation engines that combine

collaborative filtering based on user-item interaction history with content-based filtering that

uses movie metadata (Introduction to Recommender Systems, Deep Learning, n.d.).

Similar research efforts have also leveraged modern language models to improve

recommendation systems. Netflix has published studies highlighting the effectiveness of hybrid

recommender systems that use natural language embeddings alongside user behavior data

(Netflix, 2024). Additionally, platforms such as HuggingFace provide tutorials on fine-tuning

BERT for classification tasks using movie-related datasets like IMDb reviews, showcasing the

growing integration of transformer models in applied recommendation systems (Text

Classification, n.d).

Machine Learning Method Research

Various machine learning and natural language processing (NLP) techniques have been

developed to address tasks like genre classification and movie recommendation. A commonly

used baseline approach is Term Frequency – Inverse Document Frequency (TF-IDF), this is a

statistical method introduced to evaluate the importance of words within a document relative

to a corpus (Salton & Buckley, 1988). TF-IDF creates sparse vector representations based on

word frequency, allowing for similarity comparisons using metrics such as cosine similarity. In

the context of movie recommendation, TF-IDF can be used to compare plot descriptions and

retrieve similar content. However, it struggles to capture semantic relationships or contextual

meaning, which can lead to misinterpretations when different vocabularies are used to describe

similar themes. Prior studies have shown that TF-IDF fails to relate movies like The Dark Knight

and Joker or Interstellar and The Martian despite shared themes (Tamanna, 2023).

177

Internal