Visualizing multidimensional data similarities: improvements and applications
Renato Rodrigues Oliveira da Silva · 2017
Multidimensional data is increasingly more prominent and important in many application domains. Such data typically consist of a large set of elements, each of which described by several measurements (dimensions). During the design of techniques and tools to process this data, a key component is to gather insights into their structure and patterns, which can be described by the notion of similarity between elements. Among these techniques, multidimensional projections and similarity trees can effectively capture similarity patterns and handle a large number of data elements and dimensions. However, understanding and interpreting these patterns in terms of the original data dimensions is still hard. This thesis addresses the development of visual explanatory techniques for the easy interpretation of similarity patterns present in multidimensional projections and similarity trees, by several contributions. First, we propose methods that make the computation of similarity trees efficient for large datasets, and also enhance its visual representation to allow the exploration of more data in a limited screen. Secondly, we propose methods for the visual explanation of multidimensional projections in terms of groups of similar elements. These are automatically annotated to describe which dimensions are more important to define their notion of group similarity. We show next how these explanatory mechanisms can be adapted to handle both static and time-dependent data. Our proposed techniques are designed to be easy to use, work nearly automatically, and are demonstrated on a variety of real-world large data obtained from image collections, text archives, scientific measurements, and software engineering.