Guo Yuejun, Xu Qing, Sbert Mateu
Department of Informàtica i Matemàtica Aplicada, University of Girona, 17071 Girona, Spain.
School of Computer Science and Technology, Tianjin University, Tianjin 300350, China.
Entropy (Basel). 2018 Mar 2;20(3):159. doi: 10.3390/e20030159.
Analyzing trajectory data plays an important role in practical applications, and clustering is one of the most widely used techniques for this task. The clustering approach based on information bottleneck (IB) principle has shown its effectiveness for trajectory data, in which a predefined number of the clusters and an explicit distance measure between trajectories are not required. However, presenting directly the final results of IB clustering gives no clear idea of both trajectory data and clustering process. Visual analytics actually provides a powerful methodology to address this issue. In this paper, we present an interactive visual analytics prototype called IBVis to supply an expressive investigation of IB-based trajectory clustering. IBVis provides various views to graphically present the key components of IB and the current clustering results. Rich user interactions drive different views work together, so as to monitor and steer the clustering procedure and to refine the results. In this way, insights on how to make better use of IB for different featured trajectory data can be gained for users, leading to better analyzing and understanding trajectory data. The applicability of IBVis has been evidenced in usage scenarios. In addition, the conducted user study shows IBVis is well designed and helpful for users.
分析轨迹数据在实际应用中起着重要作用，而聚类是用于此任务的最广泛使用的技术之一。基于信息瓶颈（IB）原理的聚类方法已证明其对轨迹数据的有效性，其中不需要预先定义的聚类数量和轨迹之间明确的距离度量。然而，直接呈现IB聚类的最终结果并不能让人清楚地了解轨迹数据和聚类过程。可视化分析实际上提供了一种强大的方法来解决这个问题。在本文中，我们提出了一个名为IBVis的交互式可视化分析原型，以对基于IB的轨迹聚类进行富有表现力的研究。IBVis提供了各种视图，以图形方式呈现IB的关键组件和当前的聚类结果。丰富的用户交互促使不同视图协同工作，从而监控和引导聚类过程并完善结果。通过这种方式，可以为用户提供关于如何针对不同特征的轨迹数据更好地使用IB的见解，从而更好地分析和理解轨迹数据。IBVis的适用性已在使用场景中得到证明。此外，所进行的用户研究表明IBVis设计良好且对用户有帮助。
