Movie recommendation model based on probabilistic matrix decomposition using hybrid AdaBoost integration

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Recommendation systems are crucial for streaming platforms.
• Probabilistic Matrix Factorization (PMF) faces challenges with generalization and accuracy.
• Existing models like FCM-PMF, Bagging-BP-PMF, and AdaBoost-SVM-PMF have limitations.

Purpose of the Study:

• To propose a Hybrid AdaBoost Ensemble Method for improved recommendation system performance.
• To address the poor generalization ability and low prediction accuracy of the PMF model.
• To enhance user-item scoring prediction accuracy and model stability.

Main Methods:

• Utilizing fuzzy clustering with membership functions and cluster centers to compute user-item scoring matrices.
• Training scoring matrices with neural networks for enhanced prediction accuracy.
• Implementing the AdaBoost ensemble method with neural networks as base learners, using voting for final score prediction.

Main Results:

• The proposed Hybrid AdaBoost Ensemble Method demonstrates improved performance on MovieLens and FilmTrust datasets.
• Compared to the Bagging-BP-PMF model, the proposed method shows a slight increase in Mean Absolute Error (1.24% and 0.79%) and Root-Mean-Square Error (2.55% and 1.87%).
• Weighting neural network-based learners enhances model stability and score prediction universality.

Conclusions:

• The Hybrid AdaBoost Ensemble Method offers a robust approach to recommendation systems.
• The integration of fuzzy clustering, neural networks, and AdaBoost effectively improves prediction accuracy and stability.
• The method's universality is confirmed through its successful application and weighted learner approach.