Scholarlink

<p>The personalized recommendation systems are important in increasing user engagement and discovery of content in the contemporary digital platforms. In this paper, we suggest the MRGE-PRS, a Multi-Relational Graph Enhanced Personalized Recommendation System, which is intended to produce context and adaptive recommendation in the heterogeneous data setting. The framework additionally proposes complex user-item-context relations as structured heterogeneous knowledge graph, with user nodes encoding demographic and behavioral factors, item nodes encoding categorical/descriptive factors, and contextual nodes encoding time and place-based information. Multi-relational edges combine different forms of interaction cues, like ratings, reviews, as well as implicit feedback like clicks. The architecture has the latent interaction-aware representation learning and multi-level personalized attention mechanism which refine the node embeddings and dynamically focus the importance on various relational interactions. Temporal knowledge graph with an exponential decay is presented to balance between long-term user preferences and recent behavioral trends and allow modeling preferences adaptively. Besides, textual review semantics improvement also improves feature representation, whereas multi-hop reasoning reflects the dependency indirectly across the knowledge graph structure. The contextual indicators such as time-dependent trends and patterns of interaction frequency more contribute to the real-time flexibility of the recommendation process. The framework proposed will overcome most of the problematic areas of the recommendation system as it is meant to deal with such issues as data sparseness, preference drift, and heterogeneous interaction modeling. The benchmark data of MovieLens, Yelp, and Dianping where experimental assessments have been undertaken show that MRGE-PRS outperforms the current baseline techniques. Optimally configured, the model has a AUC and F1-score of 0.99, indicating that it can give scalable, robust, and context-sensitive personalized recommendations in a broad spectrum of applications.</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

Multi-relational knowledge graph-based evolutionary multi-level personalized attention GNN for recommendation systems

  • 期刊论文

  • 发表日期：

    2026年4月24日

查看全文
  • Subhankar Guha,
  • Bhramara Bar Biswal,
  • Anirban Mitra

摘要

The personalized recommendation systems are important in increasing user engagement and discovery of content in the contemporary digital platforms. In this paper, we suggest the MRGE-PRS, a Multi-Relational Graph Enhanced Personalized Recommendation System, which is intended to produce context and adaptive recommendation in the heterogeneous data setting. The framework additionally proposes complex user-item-context relations as structured heterogeneous knowledge graph, with user nodes encoding demographic and behavioral factors, item nodes encoding categorical/descriptive factors, and contextual nodes encoding time and place-based information. Multi-relational edges combine different forms of interaction cues, like ratings, reviews, as well as implicit feedback like clicks. The architecture has the latent interaction-aware representation learning and multi-level personalized attention mechanism which refine the node embeddings and dynamically focus the importance on various relational interactions. Temporal knowledge graph with an exponential decay is presented to balance between long-term user preferences and recent behavioral trends and allow modeling preferences adaptively. Besides, textual review semantics improvement also improves feature representation, whereas multi-hop reasoning reflects the dependency indirectly across the knowledge graph structure. The contextual indicators such as time-dependent trends and patterns of interaction frequency more contribute to the real-time flexibility of the recommendation process. The framework proposed will overcome most of the problematic areas of the recommendation system as it is meant to deal with such issues as data sparseness, preference drift, and heterogeneous interaction modeling. The benchmark data of MovieLens, Yelp, and Dianping where experimental assessments have been undertaken show that MRGE-PRS outperforms the current baseline techniques. Optimally configured, the model has a AUC and F1-score of 0.99, indicating that it can give scalable, robust, and context-sensitive personalized recommendations in a broad spectrum of applications.