面向拓扑感知的层次结构信息可视探索方法  

Visual Exploration Method of Hierarchical Structure Information for Topology Awareness

作　　者：谭博友 韩永国[1] 王桂娟 赵韦鑫 周锐 蔡梦杰 吴亚东 TAN Bo-you;HAN Yong-guo;WANG Gui-juan;ZHAO Wei-xin;ZHOU Rui;CAI Meng-jie;WU Ya-dong(School of Computer Science&Technology,Southwest University of Science&Technology,Mianyang 621000,China;School of Compuster Science&Engineering,Sichuan University of Science and Engineering,Zigong 643000,China)

机构地区：[1]西南科技大学计算机科学与技术学院,四川绵阳621000 [2]四川轻化工大学计算科学与工程学院,四川自贡643000

出　　处：《计算机技术与发展》2022年第11期81-87,共7页Computer Technology and Development

基　　金：国家自然科学基金(61872304,61802320)。

摘　　要：在有限的屏幕范围内,用户从有分支拥挤和节点遮蔽的层次可视化视图中获取拓扑结构信息具有挑战性。针对以上难点,提出了一种面向拓扑感知的层次结构信息探索框架。为提高用户探索拓扑结构信息的效率,提出采用重要节点评估算法。通过对以重要节点为根的子结构以视觉编码的形式进行隐藏,同时确保在保留较多的结构信息的条件下解决了分支拥挤节点遮蔽等问题。基于文本关键词的思想,定义了一种层次数据关键子结构的提取方法,通过提取关键子结构对整体拓扑结构信息进行概要,帮助用户理解整体拓扑结构特征。为提高用户对相似子结构的探索对比分析的效率,基于图表示学习算法将层次结构的节点进行向量化表示,通过将节点向量进行高斯混合聚类来构建相似子结构集合,然后采用图核计算子结构的相似度分数,通过相似度分数排序后完成相似子结构的提取。基于以上算法,设计了一个交互式的可视分析系统。通过可视分析系统完成了两项案例分析和两项用户实验,证明了所提框架的有效性。In a limited screen range,it is challenging for users to obtain topology information from hierarchical visualization views with branch congestion and node shadowing.In view of the above difficulties,a hierarchical information exploration framework for topology awareness is proposed.In order to improve the efficiency of users in exploring topology information,an important node evaluation algorithm is proposed.The sub-structure rooted in important nodes is hidden in the form of visual coding,while ensuring that the problems such as branch crowded node masking are solved under the condition of retaining more structural information.Based on the idea of text keywords,a method for extracting the key substructure of hierarchical data is defined,which summarizes the overall topology information by extracting the key substructure to help users understand the characteristics of the overall topology.In order to improve the efficiency of users’exploration and comparative analysis of similar substructure,the nodes of hierarchical structure are represented by vectorization based on graph representation learning algorithm,and the set of similar substructure is constructed by Gaussian mixture clustering of node vectors.Then the similarity score of the substructure is calculated by using the graph kernel,and the similarity substructure is extracted after the similarity score is sorted.Based on the above algorithms,an interactive visual analysis system is designed.Two case studies and two user experiments are completed through the visual analysis system to prove the effectiveness of the proposed framework.

关 键 词：层次结构数据 重要节点评估 关键子结构提取 层次结构向量化 可视分析 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]