Scholarlink

<p>Studies on college drinking—a behavior associated with health risks and reduced productivity—often rely on broadly defined peer or friendship networks. Yet, friendship networks can be further divided into more specific relational types, and the associations between such diverse social ties and drinking ties remain poorly understood. This study adopts a multilayer network framework to conduct a fine-grained examination of ten distinct types of social networks (i.e., layers)—including friendship, leadership, emotional support, and perceived drinking—by analyzing their structural similarity across three levels (node, link, and triad) and along a relationship-axis spectrum. We cluster networks based on these multi-level similarities, identifying three primary clusters (Affiliation, Leadership, and Drinking) and assess whether these structurally coherent clusters contribute to improved link prediction performance in perceived drinking networks. Both <i>k</i>-means clustering based on multi-level structural similarity and projection-based analysis along the hierarchical–horizontal spectrum revealed that perceived drinking nominations are structurally closest to leadership layers, followed by emotional support layers within the Affiliation cluster. Assessing whether these functional clusters can improve link prediction in perceived drinking networks, we find that layers within the same cluster yield predictive performance close to that of models using all layers. Notably, emotional support layers, which may reflect their structural proximity and distributed connectivity, offer the highest link prediction accuracy for drinking ties. Taken together, these findings demonstrate that coherent clusters of fine-grained, functionally and structurally aligned social layers facilitate more efficient inference in sparse or partially observed drinking-related networks. This, in turn, highlights their potential utility in predicting and preventing health-risk behaviors—such as alcohol use—that are typically difficult to observe or measure directly, and clarifies the types of relationships most strongly associated with such behaviors.</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

Functional and structural clustering of social relationship layers among college students for link prediction with applications to perceived drinking networks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月24日

查看全文
  • Eun Lee,
  • Ovidia Stanoi,
  • Xie He,
  • Yoona Kang,
  • Mia Jovanova,
  • Amanda L. McGowan,
  • David M. Lydon-Staley,
  • Dani S. Bassett,
  • Kevin N. Ochsner,
  • Zachary M. Boyd,
  • Emily B. Falk,
  • Peter J. Mucha

摘要

Studies on college drinking—a behavior associated with health risks and reduced productivity—often rely on broadly defined peer or friendship networks. Yet, friendship networks can be further divided into more specific relational types, and the associations between such diverse social ties and drinking ties remain poorly understood. This study adopts a multilayer network framework to conduct a fine-grained examination of ten distinct types of social networks (i.e., layers)—including friendship, leadership, emotional support, and perceived drinking—by analyzing their structural similarity across three levels (node, link, and triad) and along a relationship-axis spectrum. We cluster networks based on these multi-level similarities, identifying three primary clusters (Affiliation, Leadership, and Drinking) and assess whether these structurally coherent clusters contribute to improved link prediction performance in perceived drinking networks. Both k-means clustering based on multi-level structural similarity and projection-based analysis along the hierarchical–horizontal spectrum revealed that perceived drinking nominations are structurally closest to leadership layers, followed by emotional support layers within the Affiliation cluster. Assessing whether these functional clusters can improve link prediction in perceived drinking networks, we find that layers within the same cluster yield predictive performance close to that of models using all layers. Notably, emotional support layers, which may reflect their structural proximity and distributed connectivity, offer the highest link prediction accuracy for drinking ties. Taken together, these findings demonstrate that coherent clusters of fine-grained, functionally and structurally aligned social layers facilitate more efficient inference in sparse or partially observed drinking-related networks. This, in turn, highlights their potential utility in predicting and preventing health-risk behaviors—such as alcohol use—that are typically difficult to observe or measure directly, and clarifies the types of relationships most strongly associated with such behaviors.