Modeling Complex High Level Interactions in the Process of Visual Mining

Visual Mining refers to the human analytical process that uses visual representations of raw data and makes suitable inferences. During this analytical process, users are engaged in complex cognitive activities such as decision making, problem solving, analytical reasoning and learning. Now a days, users typically use interactive visualization tools, which we call as visual mining support tools (VMSTs), to mediate their interactions with the information present in visual representations of raw data and also to support their complex cognitive activities when performing visual mining. VMSTs have two main components: visual representation and interaction. Even though, these two components are fundamental aspects of VMSTs, the research on visual representation has received the most attention. It is still unclear how to design interactions which can properly support users in performing complex cognitive activities during the visual mining process. Although some fundamental concepts and techniques regarding interaction design have been in place for a while, many established researchers are of the opinion that we do not yet have a generalized, principled, and systematic understanding of interaction components of these VMSTs, and how interactions should be analyzed, designed, and integrated to support complex cognitive activities. Many researchers have recommended that one way to address this problem is through appropriate characterization of interactions in the visual mining process. Models that provide classifications of interactions have indeed been proposed in the visualization research community. While these models are important contributions for the visualization research community, they often characterize interactions at lower levels of human information interaction and high level interactions are not well addressed. In addition, some of these models are not designed to model user activity; rather they are most applicable for representing a system’s response to user activity and not the user activity itself. In this thesis, we address this problem through characterization of the interaction space of visual mining at the appropriate level. Our main contribution in this research is the discovery of a small set of classification criteria which can comprehensively characterize the interaction space of visual mining involving interactions with VMSTs for performing complex cognitive activities. These complex cognitive activities are modeled through visual mining episodes, a coherent set of activities consisting of visual mining strategies (VMSs). Using the classification criteria, VMSs are simply described as combinations of different values of these criteria. By considering all combinations, we can comprehensively cover the interaction space of visual mining. Our VMS interaction space model is unique in identifying the activity tier, a granularity of interactions (high level) which supports performance of complex cognitive activities through interactions with visual information using VMSTs. As further demonstration of the utility of this VMS interaction space model, we describe the formulation of an inspection framework which can provide quantitative measures for the support provided by VMSTs for complex cognitive activities in visual mining. This inspection framework, which has enabled us to produce a new simpler evaluation method for VMSTs in comparison to existing evaluation methods, is based soundly on existing theories and models. Both the VMS interaction space model and the inspection framework present many interesting avenues for further research