Scholarlink

<p>Extensive efforts have focused on extracting users’ profile attributes and network structures to predict retweets between two connected users, often overlooking the significant influence of social trust. This study explores retweet prediction for directly and indirectly connected users. We propose a novel prediction framework, GAT-GCNretweet, seamlessly integrating social trust relationships with user tweet content. Our framework consists of two modules: social trust embedding and content embedding. In the social trust embedding module, trust embedding is performed for each user in both trustor and trustee roles, considering user attributes, structure information, and historical retweet relationships. In the content embedding module, a double-BERT module is designed to achieve high-quality content embedding vectors. Experimental results demonstrate that GAT-GCNretweet outperforms the state-of-the-art, achieving an impressive F1-score of 0.783 on the Sina dataset, 0.795 on the dTwitter dataset, and 0.752 on the iTwitter dataset. Therefore, the GAT-GCNretweet model can effectively integrate social trust and tweet content to accurately predict retweet behavior for both directly and indirectly connected users.</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

Predicting retweets using social trust-aware graph neural network approach

  • 期刊论文

  • 发表日期：

    2025年10月7日

查看全文
  • Lidong Wang,
  • Tao Huang,
  • Yin Zhang,
  • Kang An,
  • Jie Yuan

摘要

Extensive efforts have focused on extracting users’ profile attributes and network structures to predict retweets between two connected users, often overlooking the significant influence of social trust. This study explores retweet prediction for directly and indirectly connected users. We propose a novel prediction framework, GAT-GCNretweet, seamlessly integrating social trust relationships with user tweet content. Our framework consists of two modules: social trust embedding and content embedding. In the social trust embedding module, trust embedding is performed for each user in both trustor and trustee roles, considering user attributes, structure information, and historical retweet relationships. In the content embedding module, a double-BERT module is designed to achieve high-quality content embedding vectors. Experimental results demonstrate that GAT-GCNretweet outperforms the state-of-the-art, achieving an impressive F1-score of 0.783 on the Sina dataset, 0.795 on the dTwitter dataset, and 0.752 on the iTwitter dataset. Therefore, the GAT-GCNretweet model can effectively integrate social trust and tweet content to accurately predict retweet behavior for both directly and indirectly connected users.