DEEP LEARNING–DRIVEN MOVIE RECOMMENDATION USING CBCF-CNN WITH TEXTUAL FEATURE VISUALIZATION

Abstract

The rapid expansion of the global video-on-demand (VoD) market, projected to exceed USD 257 billion by 2027, underscores the critical role of intelligent recommendation systems, with platforms such as Netflix reporting that over 80% of viewed content is driven by recommender engines. Despite their success, existing recommendation systems continue to suffer from challenges including data sparsity, cold-start problems, and limited content diversity, which restrict their scalability and personalization capabilities. To overcome these limitations, this research proposes a novel hybrid recommendation framework termed CBCF-CNN, which integrates Content-Based Filtering (CBF), Collaborative Filtering (CF), and Convolutional Neural Networks (CNNs) into a unified architecture. The proposed model exploits user–item interaction matrices alongside deep feature extraction from movie metadata and visual content using CNNs, enabling effective learning of latent representations and semantic similarities. This deep integration mitigates cold-start issues, enhances recommendation diversity, and improves personalization even under sparse data conditions. Furthermore, the parallelized CNN-based architecture supports real-time inference with low latency, making the model suitable for large-scale deployment. Overall, CBCF-CNN delivers improved accuracy, scalability, and adaptability compared to traditional recommendation approaches