Scholarlink

<p>Social media platforms are powerful tools for information dissemination, but they can also be exploited by malicious actors using deceptive accounts. Detecting these accounts is crucial to mitigating their harmful impact. In this research, we develop a classification method to classify these accounts based on their posting behavior recorded over many days. We represent daily posting activity using sequences of binary values and employ functional data analysis within a hierarchical framework. The proposed methodology identifies days when the account is active (at least one post) or inactive (no posts) in a given day and uses the posting activity during the active days to extract the account-specific latent daily posting behavior for classification. Our method is computationally efficient, provides interpretable results, and accurately discriminates between types of accounts, when applied to the Twitter data. Furthermore, our methodology is generalizable and demonstrates strong performance compared to existing alternatives in simulation studies. The package implementing the method is available on GitHub at: <a href="https://github.com/acweisha/gFPCAClassif">https://github.com/acweisha/gFPCAClassif</a>.</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

Classification of social media users via generalized multilevel functional data analysis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月13日

查看全文
  • Anthony Weishampel,
  • Ana-Maria Staicu,
  • William Rand

摘要

Social media platforms are powerful tools for information dissemination, but they can also be exploited by malicious actors using deceptive accounts. Detecting these accounts is crucial to mitigating their harmful impact. In this research, we develop a classification method to classify these accounts based on their posting behavior recorded over many days. We represent daily posting activity using sequences of binary values and employ functional data analysis within a hierarchical framework. The proposed methodology identifies days when the account is active (at least one post) or inactive (no posts) in a given day and uses the posting activity during the active days to extract the account-specific latent daily posting behavior for classification. Our method is computationally efficient, provides interpretable results, and accurately discriminates between types of accounts, when applied to the Twitter data. Furthermore, our methodology is generalizable and demonstrates strong performance compared to existing alternatives in simulation studies. The package implementing the method is available on GitHub at: https://github.com/acweisha/gFPCAClassif.