Guided Visual Analytics for Decision Making under Uncertainty

Abstract

Visual Analytics (VA) has emerged from the need to optimize decision making by involving human reasoning in sense making. The development of VA has been facilitated by significant technological advances in modern computer graphics and data processing capabilities. Involving humans in the loop aims to address high-risk scenarios where artificial intelligence (AI) automated approaches are insufficient. One active area of research with VA is the development of methods that enable the user to make efficient and effective decisions under high uncertainty. Yet, the field of VA research has not fully understood how user attitude, namely trust and confidence, interplay in VA decision making under uncertainty. Properties of the user attitude play a crucial role in optimizing VA decision making, but they are challenging to externalize and evaluate. For instance, user confidence in their decision emerges as an important indicator of effectiveness when the correctness of the decision cannot be measured. In this dissertation, we explore the use of guidance techniques to address uncertainties in VA decision making, focusing on scenarios where the correctness of decisions cannot be definitively established. Throughout this work, we learned that a multidimensional guidance mechanism can address uncertainties more effectively when uncertainties are challenging to quantify and visualize, especially in the case of subjective uncertainty. However, evaluating the effectiveness of guidance approaches requires a more comprehensive analysis of the interplay between trust and confidence within the sense-making process. Using provenance networks and SNA metrics can provide a more reliable and comprehensive assessment of user confidence, indicating that such approaches can be employed to support co-adaptive guidance.