Knowledge-Reinforced Cross-Domain Recommendation.

Area of Science:

• Artificial Intelligence
• Recommender Systems
• Data Science

Background:

• Cross-domain recommendation aims to address the cold-start problem by transferring user preferences.
• Existing methods struggle when overlapping users between domains are insufficient, impacting accuracy.
• Knowledge graphs (KGs) offer auxiliary information to bridge domain gaps.

Purpose of the Study:

• To propose a novel knowledge-reinforced cross-domain recommendation (KR-CDR) method.
• To leverage KGs for enhanced preference transfer in low-overlap scenarios.
• To improve the accuracy and performance of cross-domain recommenders.

Main Methods:

• Constructing a cross-domain knowledge graph (CDKG) from source and target domain KGs.
• Employing reinforcement learning (RL) with meta-learning on CDKG to discover meta-paths.
• Generating meta-path aggregated embedding vectors for cold-start users.

Main Results:

• The KR-CDR method effectively utilizes meta-paths derived from CDKG.
• Obtained user embeddings enable accurate rating predictions.
• Experimental results demonstrate superior performance over state-of-the-art methods on five real-world datasets.

Conclusions:

• The proposed KR-CDR method successfully addresses limitations of existing approaches.
• Leveraging KGs and RL enhances cross-domain recommendation accuracy, especially for cold-start users.
• KR-CDR offers a robust solution for improving recommender system performance in sparse data settings.