Scholarlink

<p>In recent years, graph-based recommender systems have shown remarkable progress, particularly through the integration of graph neural networks (GNNs) and auxiliary information sources such as knowledge graphs (KGs). Despite these advances, many models still face limitations in effectively fusing collaborative signals and fine-grained semantics. To address these gaps, this paper introduces KGCMC, a novel framework that leverages graph convolutional matrix completion (GCMC) to model user–item interactions, while integrating a personalized knowledge-aware attention sub-network for semantic enrichment. KGCMC jointly captures structural and semantic patterns, producing robust and context-aware embeddings for recommendation tasks. Experimental results across three benchmark datasets show competitive performance and consistent improvements over several state-of-the-art in both recall and F1-score, affirming the benefits of combining GCMC with personalized KG reasoning for recommendation.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

KGCMC: a hybrid approach combining graph convolutional matrix completion and knowledge-aware attention for recommender systems

  • 期刊论文

  • 发表日期：

    2026年5月20日

查看全文
  • Rasoul Hassanzadeh,
  • Vahid Majidnezhad,
  • Bahman Arasteh

摘要

In recent years, graph-based recommender systems have shown remarkable progress, particularly through the integration of graph neural networks (GNNs) and auxiliary information sources such as knowledge graphs (KGs). Despite these advances, many models still face limitations in effectively fusing collaborative signals and fine-grained semantics. To address these gaps, this paper introduces KGCMC, a novel framework that leverages graph convolutional matrix completion (GCMC) to model user–item interactions, while integrating a personalized knowledge-aware attention sub-network for semantic enrichment. KGCMC jointly captures structural and semantic patterns, producing robust and context-aware embeddings for recommendation tasks. Experimental results across three benchmark datasets show competitive performance and consistent improvements over several state-of-the-art in both recall and F1-score, affirming the benefits of combining GCMC with personalized KG reasoning for recommendation.