Enhancing neural collaborative filtering using hybrid feature selection for recommendation.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Online transactions have surged, increasing demand for sophisticated recommender systems.
• Current deep learning models often fail to capture comprehensive user-item correlations by focusing on either global or local interactions.
• Effective modeling of user-item interactions is crucial for personalized online services.

Purpose of the Study:

• To propose a novel deep collaborative recommendation system.
• To effectively capture both local and global higher-order interactions between users and items.
• To improve the performance and prevent overfitting in recommender systems.

Main Methods:

• Developed a deep collaborative recommendation system utilizing a convolutional neural network.
• Incorporated an outer product matrix and a hybrid feature selection module to capture diverse interactions.
• Integrated generalized matrix factorization weights for network optimization and regularization.

Main Results:

• The proposed system successfully captures local and global higher-order user-item interactions.
• Experiments on two real-world datasets demonstrated superior performance compared to baseline methods.
• The approach proved effective across datasets with varying sparsity levels.

Conclusions:

• The novel deep collaborative recommendation system effectively models complex user-item correlations.
• The integration of convolutional neural networks and advanced feature selection enhances recommendation accuracy.
• This method offers a significant advancement for personalized online services and recommender system research.