|
|
 |
| |
|
|
|
Next-Generation Movie Recommenders: Leveraging Hybrid Deep Learning for Enhanced Personalization |
|
|
|
PP: 957-981 |
|
|
|
doi:10.18576/amis/180504
|
|
|
|
Author(s) |
|
|
|
Deema Mohammed Alsekait,
Ahmed Younes Shdefat,
Nour Mostafa,
Alaa Mohamed Mohamed Hamdy,
Hanaa Fathi,
Diaa Salama AbdElminaam,
|
|
|
|
Abstract |
|
|
| This study aims to advance movie recommendation systems by integrating and comparing multiple machine learning and deep learning algorithms, specifically focusing on Contact-Based Filtering and Content Filtering techniques such as TF-IDF, Decision Trees, K-Nearest Neighbors, Singular Value Decomposition (SVD), Neural Collaborative Filtering (NCF), ALS, and the AUTOENCODER. We comprehensively evaluated these methodologies using three distinct datasets—MovieLens Dataset, The Movies Dataset, and TMDB Movie Dataset. The results indicate that different algorithms excel depending on the dataset characteristics. SVD shows superior performance on the MovieLens Dataset, AUTOENCODER excels in The Movies Dataset, and a hybrid approach proves most effective on the TMDB Movie Dataset. The study emphasizes the importance of considering dataset specificities when selecting recommendation algorithms to optimize system performance. Our findings contribute to developing more personalized, accurate movie recommendation systems and highlight the potential of hybrid approaches in addressing diverse user preferences and content complexities. Integrating various machine learning techniques significantly enhances the adaptability and accuracy of recommendation systems, presenting a substantial advancement in personalized content delivery. We leverage three datasets (MovieLens Dataset, The Movies Dataset, and TMDB Movie Dataset) for comprehensive evaluation and comparison. In our analyses, SVD demonstrated superior performance in the MovieLens 20M Dataset, while transitioning to The Movies Dataset highlighted AUTOENCODER’s exceptional performance, and the TMDB 5000 Movie Dataset showcased the hybrid recommender as the most effective. The study emphasizes the importance of considering specific dataset characteristics for optimal recommendation algorithm selection. Integrating Contact- Based Filtering, Content Filtering, and Neural Collaborative Filtering significantly improves recommendation system accuracy. While Singular Value Decomposition excels, further investigation considering dataset characteristics and user preferences is deemed essential for algorithm selection in diverse scenarios.
|
|
|
|
|
|
|
|
