Dual-adaptive imputation graph neural network for knowledge-aware recommendation.

Area of Science:

• Artificial Intelligence
• Computer Science
• Data Science

Background:

• Recommender systems personalize content delivery for enhanced user experience.
• Knowledge graph-based systems address data sparsity but struggle with user-item matrix utilization and heterogeneous signal integration.

Purpose of the Study:

• To propose DAIGNN (Dual-Adaptive Imputation Graph Neural Network), a novel framework to overcome limitations in current recommender systems.
• To improve the utilization of user-item interaction data and integrate collaborative information with heterogeneous knowledge graph signals.

Main Methods:

• Developed a similarity-driven imputation mechanism to create an Imputation Graph, reducing data sparsity.
• Incorporated multi-source auxiliary information for richer contextual and relational semantics.
• Implemented a dual-adaptive feature fusion mechanism for dynamic integration of heterogeneous graph information.

Main Results:

• DAIGNN demonstrated superior effectiveness on four real-world datasets.
• Achieved an average improvement of 3.1% in AUC and 2.0% in F1-score over state-of-the-art baselines.
• Confirmed robustness across diverse recommendation settings.

Conclusions:

• DAIGNN effectively addresses key limitations in knowledge graph-based recommender systems.
• The proposed framework enhances personalization and recommendation accuracy.
• DAIGNN offers a robust solution for improving user experience in recommender systems.