Hybrid recommendation algorithm based on real-valued RBM and CNN.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• The proliferation of big data necessitates advanced information retrieval methods.
• Traditional recommendation systems, like those using limited Boltzmann machines for collaborative filtering, often overlook crucial data such as item content and user characteristics.
• This limitation impacts the accuracy of predicted scores.

Purpose of the Study:

• To propose a modified hybrid recommendation algorithm that addresses the limitations of existing methods.
• To enhance recommendation accuracy by integrating diverse data sources, including textual information.
• To leverage the Gaussian restricted Boltzmann machine for improved rating prediction.

Main Methods:

• Textual data from users and items are converted into numerical vectors using an embedding layer.
• Convolutional neural networks (CNNs) extract latent features from this text data.
• These latent features are combined with rating vectors to create comprehensive user and item vectors.
• A Gaussian restricted Boltzmann machine (GRBM) is employed for rating prediction, utilizing a fused user-item matrix.

Main Results:

• Simulation experiments demonstrate the feasibility and effectiveness of the proposed algorithm.
• The enhanced approach shows improved performance in predicting user ratings compared to methods relying solely on scoring matrices.
• Incorporating item content and user characteristics significantly benefits recommendation accuracy.

Conclusions:

• The modified hybrid recommendation algorithm based on GRBM offers a robust solution for accurate rating prediction.
• Integrating textual content and user features enhances the capabilities of collaborative filtering.
• This research contributes a feasible and effective method for improving recommendation systems in big data environments.