Order, Spacing, and Clustering in Visual Exploration of Large Scale Data

本研究项目的重点是开发可视化、交互和数据管理技术，以解决大规模可视化数据挖掘中的高维和数据类型异构性问题。基本方法是在数据集的各个维度上以及在包含名义或分类值的各个维度内应用多分辨率聚类策略，并利用数据轴和数据点的排序和定位来强调数据中的关系。对于可视化，任务涉及开发确定数据和维度内和之间的排序和可变间距的方法，以及将维度聚集成多分辨率抽象，并将它们整合到几种现有的多变量显示技术中。对于交互，开发了用于在多分辨率空间内进行直观导航和探索的工具。这包括允许用户指导对数据对象和维度的集群进行拆分和分组的交互式重新集群工具。对于数据管理，任务涉及为高维数据访问开发高维索引和多分辨率数据视图管理，以及支持实时视觉探索的缓存和预取策略。评估了显示和交互工具在大数据集上的易用性和性能。这项研究的结果将使生物信息学、地球和空间科学以及电子商务等领域的数据分析人员能够以交互方式探索正在产生的日益庞大和复杂的数据集。

The focus of this research project is to develop visualization, interaction, and data management technologies to address the problems of high dimensionality and data type heterogeneity in large-scale visual data mining. The basic approach is to apply multiresolution clustering strategies across the dimensions of a data set as well as within individual dimensions containing nominal or categorical values, and exploit the ordering and positioning of data axes and data points to emphasize relationships within the data. For visualization, the tasks involve the development of methods for determining ordering and variable spacing within and between data and dimensions as well as clustering of dimensions into multi-resolution abstractions, and integrating them into several existing multivariate display techniques. For interaction, tools for intuitive navigation and exploration within the multiresolution spaces are developed. This includes interactive reclustering tools to allow users to guide the process of splitting and grouping clusters of data objects and dimensions. For data management the tasks involve the development of high-dimensional indexing and multi-resolution data view management for high-dimensional data access, and caching and prefetching strategies to support real-time visual exploration. The ease of use as well as performance of the display and interactive tools over large data sets is assessed. The results of this research will provide data analysts in domains such as bioinformatics, earth and space sciences, and e-commerce the ability to interactively explore the increasingly large and complex data sets being generated.