Neural Matrix Factorization Recommendation for User Preference Prediction Based on Explicit and Implicit Feedback.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Recommendation systems leverage user interaction data, with explicit and implicit feedback being key heterogeneous data types.
• Current deep learning models often underutilize the complementary strengths of combined feedback, frequently relying solely on binary implicit data.

Purpose of the Study:

• To propose a novel neural matrix factorization recommendation algorithm (EINMF) that effectively integrates explicit and implicit feedback.
• To enhance recommendation system performance by fully exploiting the synergistic advantages of heterogeneous user feedback data.

Main Methods:

• Utilizing neural networks to learn nonlinear features from user-item interactions across both explicit and implicit feedback.
• Integrating traditional matrix factorization with explicit feedback to capture both direct and latent user preferences.
• Designing a new loss function tailored for explicit-implicit feedback to optimize model parameters via neural network training.

Main Results:

• The proposed EINMF algorithm demonstrates superior feasibility, validity, and robustness compared to multiple baseline models.
• Performance was validated on two real-world datasets, showcasing the effectiveness of the integrated feedback approach.

Conclusions:

• The EINMF algorithm successfully integrates explicit and implicit feedback for more accurate user preference prediction.
• This approach offers a significant improvement over existing methods, paving the way for more effective personalized recommendation systems.