IMGC-GNN: A multi-granularity coupled graph neural network recommendation method based on implicit relationships.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Collaborative filtering (CF) is a key method in application recommendation.
• Existing CF methods struggle to learn attribute and interaction factors separately, limiting understanding of user behavior.
• This leads to suboptimal performance in personalized recommendations.

Purpose of the Study:

• To propose a novel multi-granularity coupled graph neural network recommendation method (IMGC-GNN).
• To address the limitations of existing CF methods by learning attribute and interaction factors distinctly.
• To enhance application recommendation accuracy by incorporating contextual information.

Main Methods:

• Constructed a three-layer coupled graph incorporating user-application interactions with time and space context.
• Employed graph neural networks to learn attribute factors by decomposing the graph into user, application, and context homogeneous graphs.
• Developed interaction representation learning using a homogeneous graph of user-context-application interactions, leveraging node and structural similarity.

Main Results:

• The proposed IMGC-GNN method demonstrated superior performance compared to seven baseline methods.
• Experiments on real-world data from three cities validated the model's effectiveness.
• The method achieved the best performance in top-k recommendations, indicating accurate user targeting.

Conclusions:

• IMGC-GNN successfully learns attribute and interaction factors separately, offering deeper insights into user behavior.
• The integration of contextual information significantly improves the accuracy of application recommendations.
• The model provides a robust framework for personalized recommendations in diverse contexts.