A transformer-based architecture for collaborative filtering modeling in personalized recommender systems.

Area of Science:

• Artificial Intelligence
• Recommender Systems
• Machine Learning

Background:

• Recommender systems are vital for personalized content delivery in e-commerce, social media, and entertainment.
• Accurate modeling of user preferences is crucial for movie recommendation systems to boost user satisfaction.
• Artificial Intelligence (AI) is increasingly used to enhance the precision and adaptability of these systems.

Purpose of the Study:

• To propose a novel transformer-based architecture, MetaBERTTransformer4Rec (MBT4R), for movie recommendation.
• To demonstrate the superiority of MBT4R over existing state-of-the-art methods.
• To improve the accuracy and personalization of movie recommendations for enhanced user satisfaction.

Main Methods:

• Developed a novel transformer-based architecture, MetaBERTTransformer4Rec (MBT4R).
• Utilized a self-attention mechanism to capture sequential dependencies and contextual relationships.
• Conducted empirical analysis on two MovieLens datasets.

Main Results:

• MBT4R achieved the lowest RMSE (0.62) and MAE (0.45), and the highest R² (0.39).
• The model significantly outperformed traditional machine learning, matrix factorization, and deep learning benchmarks.
• Demonstrated superior performance compared to DT, KNN, RF, XGB, SVD, and GRU models.

Conclusions:

• AI techniques, specifically the MBT4R model, effectively enhance recommendation system accuracy and personalization.
• The proposed model provides a pathway for future advancements in personalized user experiences.
• Accurate prediction of user preferences leads to tailored film suggestions and increased user satisfaction.