Visual analytics of dynamic higher order information

Abstract

Listen

Translate article

Nearest neighbour query and nearest region query providing higher-order information are the two most popular queries with spatio-temporal data. These types of higher-order information serve as the basis for what-if and what-happen scenarios. As our world turns into a more complex and uncertain environment with different emerging technologies, one of the challenging issues many decision-makers face is a data-rich, information-rich, but knowledge-poor situation. Multiple and dynamic datasets have been collected by various devices (sources). However, the time variant nature of these dynamic datasets will require different analytic tools and techniques to pre-process and analyse. Understanding and gaining insights into dynamic spatio-temporal data is of great importance in current ubiquitous environments, and visual analytics is a solid candidate to amplify user's ability to detect expected patterns and to discover unexpected patterns from this type of dynamic spatio-temporal data. Previous what-if analysis research mainly focused on single (univariate) datasets only. Existing tools that work with univariate spatio-temporal data do not have the data structure and cannot effectively support trajectory datasets. Moreover, related visual analytics methods to deal with dynamic spatio-temporal data, especially for higher-order information, are very limited. These methods are usually non-interactive that cannot support data exploration for users. The visualisation techniques used in existing methods also mainly focused on quantitative information but not qualitative information. In terms of different types of information, previous studies mostly focused on geometrical information with very limited attention in topological and directional information. There is also a lack of measurements and data mining techniques for decision makers to compare similarity/dissimilarity of higher-order information of trajectories. The overall aim of this thesis is to investigate the application of visual analytics tools to provide various higher-order information with multivariate datasets of moving objects to provide decision support for users. First, we will design a robust data structure that can effectively support multivariate Point of Interests targets and multivariate dynamic moving objects of higher-order information. Second, the data structure needs to support not only the most widely used qualitative/quantitative geometrical higher-order information, but topological and directional information. Third, we will design and develop various visual analytical tools that help users to gain insights into the higher-order nature of trajectory data. Fourth, the proposed tools will be evaluated against effectiveness and efficiency. Fifth, we will define several dissimilarity measures for geometrical/topological/directional higher-order information and related data mining algorithms that return top-k similar trajectories. Lastly, we will present case studies with real data to validate the developed data structure and visual analytics tools. The outcomes achieved by this research provide a number of contributions to the greater body of knowledge in data analytics and spatial temporal data analysis. We propose a set of visual analytics tools for exploring and analysing multivariate Points of Interest targets and multivariate dynamic moving objects of higher-order information. The contributions of this research include a new vector-based unified data structure, which allows datasets between Points of Interest and trajectories to be modelled and quantified. Two ways of data preprocessing of trajectory can be addressed when data is too dense or too sparse. In addition, there are five different types of visual analytics tools and three types of new measurement metrics for describing and comparing the similarity of higher-order information. These include the visual analytical tools, namely, higher-order parallel coordinates, higher-order spider, higher-order radar, higher-order treemaps, and higher-order DNA. For the measurement metrics, this research introduces three metrics, namely directional dissimilarity measurement, parallel coordinate multivariate measurement, and higher-order DNA impact factor. These visual tools and measurement metrics address the limitation of existing analytical tools to provide interactive tools for users to explore both qualitative and quantitative multivariate higher-order information with integration of all three different types of higher-order information. A real-world emergency management case study will demonstrate the usefulness and applicability of these new visual analytics tools.