Interdependent-path Recurrent Embedding for Knowledge Graph-aware Recommendation

• Published in: Journal of Applied Science and Engineering Vol. 29; no. 3; pp. 531 - 543
• Main Authors: Xiao Sha, Jianwen Wang, Xiaoran Xu, Jianchuan Ding
• Format: Journal Article
• Language: English
• Published: Tamkang University Press 2026
• Subjects: attention mechanism; collaborative filtering; knowledge graphs; recommender systems
• DOI: 10.6180/jase.202603_29(3).0004
• ISSN: 2708-9967

Knowledge graphs (KGs) have demonstrated their effectiveness in providing high-quality recommendations by incorporating rich semantic relationships between entities. However, existing KG-aware recommendation methods face significant challenges in sufficiently exploiting both the structural and semantic information while maintaining computational efficiency. We propose the Interdependent-path Recurrent Embedding (IPRE) framework that addresses these limitations through novel interdependent path construction and attentive encoding. The framework automatically generates interdependent paths connecting user-item pairs, preserving both semantic relationships and topological dependencies with linear time complexity. A dedicated attentive recurrent network then encodes these paths by learning relation-aware representations and adaptively weighting different predecessors’ influence. Comprehensive experiments on three real-world datasets demonstrate IPRE’s superiority, achieving average improvements of 8.79% in Hit ratio and 9.40% in NDCG over state-of-the-art methods. The framework shows particular effectiveness in sparse data scenarios, while maintaining competitive computational efficiency. These results validate IPRE’s capability to effectively transform KG information into accurate recommendations through its innovative path modeling approach.