Scholarlink

<p>Temporal aggregation is widely used in dynamic graph visualization to reduce visual complexity and improve scalability. However, for community evolution analysis, coarse aggregation may compress identity-continuity cues that are needed to recognize structural events. We introduce Visual Aliasing Entropy (VA-Entropy), a data-derived diagnostic indicator computed from within-window community-label distributions, to quantify aggregation-induced identity compression. VA-Entropy is not derived from rendered pixels, layouts, visual saliency, or visual encodings; instead, its perceptual relevance is evaluated through user behavior. We conducted a controlled study with 31 technically trained participants and 2232 event-classification trials on synthetic dynamic networks with explicitly specified membership transitions. The results show that higher VA-Entropy is associated with lower recognition accuracy, longer response time, greater interaction effort, and lower confidence. In adjusted GEE models controlling for event type and aggregation window size, each 0.1 increase in normalized VA-Entropy was associated with approximately 34% lower odds of correct classification. These findings suggest that temporal resolution should be treated as a perceptually consequential design parameter. VA-Entropy can complement structural or visual-complexity criteria by flagging aggregation windows that appear simple but may hide identity transitions important for community evolution interpretation. Code, stimuli, anonymized behavioral logs, and analysis scripts are available in the project repository: <a href="https://github.com/jinghanbreeze-sketch/DyG-Perception/tree/v1.0.1">https://github.com/jinghanbreeze-sketch/DyG-Perception/tree/v1.0.1</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

Visual aliasing entropy for perception-aware temporal aggregation in dynamic graph visualization

  • 期刊论文

  • 发表日期：

    2026年8月5日

查看全文
  • Jinghan Bai,
  • Yiming Lin,
  • Huijie Zhang

摘要

Temporal aggregation is widely used in dynamic graph visualization to reduce visual complexity and improve scalability. However, for community evolution analysis, coarse aggregation may compress identity-continuity cues that are needed to recognize structural events. We introduce Visual Aliasing Entropy (VA-Entropy), a data-derived diagnostic indicator computed from within-window community-label distributions, to quantify aggregation-induced identity compression. VA-Entropy is not derived from rendered pixels, layouts, visual saliency, or visual encodings; instead, its perceptual relevance is evaluated through user behavior. We conducted a controlled study with 31 technically trained participants and 2232 event-classification trials on synthetic dynamic networks with explicitly specified membership transitions. The results show that higher VA-Entropy is associated with lower recognition accuracy, longer response time, greater interaction effort, and lower confidence. In adjusted GEE models controlling for event type and aggregation window size, each 0.1 increase in normalized VA-Entropy was associated with approximately 34% lower odds of correct classification. These findings suggest that temporal resolution should be treated as a perceptually consequential design parameter. VA-Entropy can complement structural or visual-complexity criteria by flagging aggregation windows that appear simple but may hide identity transitions important for community evolution interpretation. Code, stimuli, anonymized behavioral logs, and analysis scripts are available in the project repository: https://github.com/jinghanbreeze-sketch/DyG-Perception/tree/v1.0.1.