SUPPORT EFFECTIVE DISCOVERY MANAGEMENT IN VISUAL ANALYTICS

Visual analytics promises to supply analysts with the means necessary to ana- lyze complex datasets and make effective decisions in a timely manner. Although significant progress has been made towards effective data exploration in existing vi- sual analytics systems, few of them provide systematic solutions for managing the vast amounts of discoveries generated in data exploration processes. Analysts have to use off line tools to manually annotate, browse, retrieve, organize, and connect their discoveries. In addition, they have no convenient access to the important discoveries captured by collaborators. As a consequence, the lack of effective discovery manage- ment approaches severely hinders the analysts from utilizing the discoveries to make effective decisions. In response to this challenge, this dissertation aims to support effective discov- ery management in visual analytics. It contributes a general discovery manage- ment framework which achieves its effectiveness surrounding the concept of patterns, namely the results of users’ low-level analytic tasks. Patterns permit construction of discoveries together with users’ mental models and evaluation. Different from the mental models, the categories of patterns that can be discovered from data are pre- dictable and application-independent. In addition, the same set of information is often used to annotate patterns in the same category. Therefore, visual analytics sys- tems can semi-automatically annotate patterns in a formalized format by predicting what should be recorded for patterns in popular categories. Using the formalized an- notations, the framework also enhances the automation and efficiency of a variety of discovery management activities such as discovery browsing, retrieval, organization, association, and sharing. The framework seamlessly integrates them with the visual interactive explorations to support effective decision making. Guided by the discovery management framework, our second contribution lies in proposing a variety of novel discovery management techniques for facilitating the discovery management activities. The proposed techniques and framework are im- plemented in a prototype system, ManyInsights, to facilitate discovery management in multidimensional data exploration. To evaluate the prototype system, two long- term case studies are presented. They investigated how the discovery management techniques worked together to benefit exploratory data analysis and collaborative analysis. The studies allowed us to understand the advantages, the limitations, and design implications of ManyInsights and its underlying framework