Recommendation algorithm based on attributed multiplex heterogeneous network.

Area of Science:

• Deep Learning
• Network Science
• Recommender Systems

Background:

• Processing large network models with billions of nodes and numerous edge types presents significant challenges in deep learning.
• Current recommendation systems struggle with accuracy when applying large network embeddings due to these processing deficiencies.

Purpose of the Study:

• To develop a novel framework, SSN_GATNE-T, that overcomes the limitations of processing large networks for improved recommendation accuracy.
• To enhance the mining of potential user information and mitigate information loss in recommendation models.

Main Methods:

• Combines attributed multiplex heterogeneous networks with an attention mechanism incorporating softsign and sigmoid functions.
• Introduces a new framework named SSN_GATNE-T (Softsign-based Attention mechanism for Graph Attributed Network Transductive Embeddings).
• Utilizes the Adam optimizer for faster model convergence and effective tuning.

Main Results:

• SSN_GATNE-T demonstrated superior performance across ROC-AUC, PR-AUC, and F1-score metrics on Amazon and YouTube datasets compared to existing models.
• The framework effectively handles large networks with numerous nodes and edge types, improving interaction information accuracy.
• The model successfully addresses the cold start problem in recommendation systems.

Conclusions:

• The proposed SSN_GATNE-T framework significantly enhances recommendation performance by effectively processing large, complex networks.
• The integration of attributed multiplex heterogeneous networks and an improved attention mechanism leads to more accurate user-item information mining.
• SSN_GATNE-T offers a robust solution for improving recommendation accuracy and tackling the cold start issue in large-scale network embedding applications.