Scholarlink

The exponential increase in social network data necessitates the development of advanced analytical techniques to better understand user interactions. This paper presents a robust framework that employs Graph Neural Networks (GNNs) for relationship prediction, enhanced by a variety of hyperparameter optimization methods, including Grid Search, Random Search, Bayesian Optimization, Evolutionary Optimization, and Neural Architecture Search (NAS). We also integrate the Louvain method for community detection, which uncovers the underlying structures within the network.Our comprehensive evaluation reveals the considerable impact of these methodologies on model performance, with Bayesian Optimization yielding the best results. Furthermore, the community detection analysis provides valuable insights that inform targeted engagement strategies by distinguishing between influential and non-influential users. This study significantly advances both theoretical understanding and practical applications in social network analysis, offering pathways for future research aimed at refining GNN architectures and incorporating additional features. Such enhancements could further improve the model’s predictive power, ultimately fostering more meaningful interactions and community growth within social platforms.

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

Analyzing Social Networks Using Graph Neural Networks with Advanced Hyperparameter Optimization Techniques

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Thuy Thi Tran,
  • Nghia Quoc Phan

摘要

The exponential increase in social network data necessitates the development of advanced analytical techniques to better understand user interactions. This paper presents a robust framework that employs Graph Neural Networks (GNNs) for relationship prediction, enhanced by a variety of hyperparameter optimization methods, including Grid Search, Random Search, Bayesian Optimization, Evolutionary Optimization, and Neural Architecture Search (NAS). We also integrate the Louvain method for community detection, which uncovers the underlying structures within the network.Our comprehensive evaluation reveals the considerable impact of these methodologies on model performance, with Bayesian Optimization yielding the best results. Furthermore, the community detection analysis provides valuable insights that inform targeted engagement strategies by distinguishing between influential and non-influential users. This study significantly advances both theoretical understanding and practical applications in social network analysis, offering pathways for future research aimed at refining GNN architectures and incorporating additional features. Such enhancements could further improve the model’s predictive power, ultimately fostering more meaningful interactions and community growth within social platforms.