Managing Uncertainty in Visual Analytics: Designing Explanation Interfaces for Recommender Systems and Prediction Models
Visual Analytics is the science that studies the visual representation of information to enable interactive analysis of large and/or complex sets of information. These representations provide a guide to the thinking process of the decision-maker to facilitate correct interpretation of otherwise intractable data. The nature of the data, however, can heavily affect the effectiveness of guidance. Missing measurements, heterogeneous sources and errors in the data are usually present in even the most curated data. In addition, limitations of the analysis algorithm can introduce a bias. Current visualization techniques proposed by Visual Analytics fail to represent this uncertainty. The lack of awareness about data quality could lead the decision-maker to take a course of action based on flawed information. Our project will research new visualization techniques that can communicate the underlying uncertainty as estimated by data-processing experts to domain-experts that have little or no knowledge on data-processing. These visualization techniques will help domain-experts to establish their degree of trust in the presented information and to take decisions with a greater knowledge of the quality of the data. In addition, the project will research interaction techniques to capture feedback from domain- experts and to incorporate such feedback in the analysis process.