Scholarlink

<p>This study addresses key limitations in existing recommendation systems, including data sparsity, cold-start issues, lack of contextual awareness, limited diversity, and data bias. Recently, various studies have attempted to overcome these challenges by integrating advanced analytical methods and adopting state-of-the-art algorithms. However, these issues remain persistent and unresolved. To solve this, the present research proposes a hybrid approach that combines clustering analysis and social network analysis (SNA) with modern recommendation algorithms. Specifically, we evaluate the effectiveness of this approach in recommending items to new users by implementing and comparing four algorithms—FM, DeepFM, AutoInt, and DCN-V2—alongside four feature processing strategies: basic, clustering, SNA, and cluster &amp; SNA. For performance evaluation, we perform 10 tests in a configuration where users cannot be separated for training, validation, and testing. The performance of each combination is assessed using ANOVA and Tukey HSD post-hoc tests. Experimental results indicate that the integration of clustering and SNA methods contributes to partial improvements in recommendation performance. This study offers a comprehensive framework for mitigating major challenges in recommendation systems and suggests a promising direction for effective model design. It is expected to contribute to the advancement of recommendation technologies across various domains.</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

Hybrid recommendation framework: integrating behavioral clustering, network centrality, and advanced deep learning models

  • 期刊论文

  • 发表日期：

    2026年4月30日

查看全文
  • Heon Baek,
  • Eui-Bang Lee

摘要

This study addresses key limitations in existing recommendation systems, including data sparsity, cold-start issues, lack of contextual awareness, limited diversity, and data bias. Recently, various studies have attempted to overcome these challenges by integrating advanced analytical methods and adopting state-of-the-art algorithms. However, these issues remain persistent and unresolved. To solve this, the present research proposes a hybrid approach that combines clustering analysis and social network analysis (SNA) with modern recommendation algorithms. Specifically, we evaluate the effectiveness of this approach in recommending items to new users by implementing and comparing four algorithms—FM, DeepFM, AutoInt, and DCN-V2—alongside four feature processing strategies: basic, clustering, SNA, and cluster & SNA. For performance evaluation, we perform 10 tests in a configuration where users cannot be separated for training, validation, and testing. The performance of each combination is assessed using ANOVA and Tukey HSD post-hoc tests. Experimental results indicate that the integration of clustering and SNA methods contributes to partial improvements in recommendation performance. This study offers a comprehensive framework for mitigating major challenges in recommendation systems and suggests a promising direction for effective model design. It is expected to contribute to the advancement of recommendation technologies across various domains.