Iterative heterogeneous graph learning for knowledge graph-based recommendation.

Area of Science:

• Artificial Intelligence
• Data Science
• Computer Science

Background:

• Knowledge graphs (KGs) offer rich, multi-relational data valuable for recommendation systems.
• Existing methods struggle to fully leverage the heterogeneous nature of KGs, limiting recommendation accuracy.
• Capturing user interests from complex KG structures remains a challenge.

Purpose of the Study:

• To propose a novel recommendation method, Heterogeneous Graph Knowledge Recommendation (HGKR), that effectively utilizes the heterogeneous structure of KGs.
• To enhance recommendation performance by addressing the limitations of current approaches in handling multi-type nodes and relations within KGs.

Main Methods:

• Developed HGKR, a method employing iterative heterogeneous graph learning on knowledge graphs.
• Integrated graph neural networks for fine-grained modeling of entity relationships at both graph and semantic levels.
• Implemented a knowledge-perceiving item filter with an attention mechanism to better capture user preferences.

Main Results:

• HGKR demonstrated superior performance compared to benchmark models in recommendation tasks.
• The method effectively models the heterogeneous characteristics of knowledge graphs for improved recommendations.
• Experiments on two datasets confirmed the excellence of the proposed HGKR approach.

Conclusions:

• HGKR offers a significant advancement in knowledge graph-based recommendation systems.
• The proposed method provides a more effective way to model and utilize heterogeneous information for personalized recommendations.
• HGKR outperforms existing methods, highlighting the potential of heterogeneous graph learning in this domain.