Scholarlink

<p>Personalized recommendation systems are essential for enhancing user experiences on social networks, particularly in Point of Interest (POI) recommendations for tourists navigating unfamiliar regions. Existing sequential models often rely on unidirectional architectures, limiting their ability to capture comprehensive contextual information and dynamic user preferences. This paper proposes BERT-KGRec, a hybrid recommendation framework combining Bidirectional Encoder Representations from Transformers (BERT) and a knowledge graph for contextualized POI recommendations. BERT captures semantic and contextual data from user-generated posts, while the knowledge graph enriches recommendations with relational and contextual insights. Sentiment analysis of user posts further refines preference modeling, and demographic data integration addresses the cold-start problem. Experimental evaluations on real-world datasets from Flickr and Yelp demonstrate significant improvements in measurements like NDCG, F1-Score, MAP, and RMSE in contrast to cutting-edge techniques, highlighting the model’s potential in delivering accurate, personalized, and context-aware recommendations.</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

Deep hybrid framework of BERT and knowledge graph for personalized POI recommendations in social networks

  • 期刊论文

  • 发表日期：

    2025年9月23日

查看全文
  • Shuyun Chang,
  • Yang Yang

摘要

Personalized recommendation systems are essential for enhancing user experiences on social networks, particularly in Point of Interest (POI) recommendations for tourists navigating unfamiliar regions. Existing sequential models often rely on unidirectional architectures, limiting their ability to capture comprehensive contextual information and dynamic user preferences. This paper proposes BERT-KGRec, a hybrid recommendation framework combining Bidirectional Encoder Representations from Transformers (BERT) and a knowledge graph for contextualized POI recommendations. BERT captures semantic and contextual data from user-generated posts, while the knowledge graph enriches recommendations with relational and contextual insights. Sentiment analysis of user posts further refines preference modeling, and demographic data integration addresses the cold-start problem. Experimental evaluations on real-world datasets from Flickr and Yelp demonstrate significant improvements in measurements like NDCG, F1-Score, MAP, and RMSE in contrast to cutting-edge techniques, highlighting the model’s potential in delivering accurate, personalized, and context-aware recommendations.