A graph neural network framework based on preference-aware graph diffusion for recommendation.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Science

Background:

• Graph-based deep learning methods are effective for point-of-interest (POI) recommendation by analyzing user check-in data.
• Existing methods struggle to capture deep graph structural information and user-specific preferences, leading to suboptimal recommendations.

Purpose of the Study:

• To propose a novel framework, preference-aware graph diffusion (PGD), to address limitations in current graph-based recommendation systems.
• To enhance the capture of deep graph structural information and incorporate both global and user-specific preferences.

Main Methods:

• Constructing two distinct graphs representing global and user preferences.
• Applying a graph diffusion process to extract structural information and generate weighted adjacency matrices.
• Utilizing graph neural network backbones and a learnable aggregation module for adaptive representation learning.

Main Results:

• The proposed PGD framework demonstrated superior performance in POI recommendation tasks.
• Experiments on four real-world datasets confirmed the effectiveness of PGD over mainstream graph-based methods.
• PGD successfully captured deep graph structural information and learned adaptive representations.

Conclusions:

• The preference-aware graph diffusion (PGD) framework offers a significant advancement in POI recommendation.
• PGD effectively models user check-in behavior by integrating global and user preferences.
• The proposed method provides a robust approach for learning user and POI representations in recommendation systems.